AI tool management method, system and device, medium and program product

Abstract

The invention discloses an AI tool management method, system and device, a medium and a program product, and the method comprises the steps: obtaining tool metadata of an AI tool based on an MCP protocol, constructing a relation graph of a tool shelf group and the AI tool, and configuring a resource robot unit for the tool shelf group; the shelf visibility scores of the AI tools are evaluated in real time, and the AI tools are divided into an on-shelf candidate list and an off-shelf candidate list based on the shelf visibility scores; the racking candidate list is displayed to the user terminal; and allocating a target AI tool in an AI tool selection request of the user terminal to a corresponding resource robot unit for execution, and triggering an on-shelf event or an off-shelf event of the AI tool according to a resource occupation condition monitored in real time. According to the invention, automatic registration of the AI tool, adaptive resource allocation and dynamic regulation and control of tool loading and unloading can be realized, and the response real-time performance and the resource utilization rate of an AI supermarket system are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a management method, system, device, medium, and program product for AI tools. Background Technology

[0002] Currently, cloud storage platforms' "AI Supermarkets" offer various AI tools, such as "AI photo editing" and "AI meeting minutes." Existing configurations typically include steps like tool registration, delisting / removal decisions, and user access. Tool registration involves developers submitting AI tool metadata through a management interface, which is then manually added to the database by operations staff after review. Delisting / removal decisions refer to operations staff manually adding some tools to increase exposure based on operational popularity strategies, and manually removing tools when server resources are strained and their effectiveness is reduced. User access refers to users opening the "AI Supermarket" interface, where the system retrieves available tools from a fixed list and consumes server computing resources to execute tasks.

[0003] However, in the current solution, the listing and delisting of AI tools are manually configured by operations staff. This manual configuration causes the tools to lag behind changes in user demand and cannot adapt to fluctuations in resource load in real time. For example, during peak periods when computing resources are strained, tools cannot be automatically delisted, resulting in service degradation. Furthermore, with the surge in the number of AI tools, manual management is inefficient and cannot support large-scale dynamic registration. Summary of the Invention

[0004] The purpose of this invention is to provide a management method, system, device, medium, and program product for AI tools. By introducing the MCP (Model Context Protocol) protocol to build a standardized tool context communication mechanism, it enables automated registration of AI tools, adaptive resource allocation, and dynamic control of tool listing and delisting, thereby improving the responsiveness and resource utilization of the AI ​​supermarket system.

[0005] To achieve the above objectives, embodiments of the present invention provide a method for managing AI tools, including: Based on the MCP protocol, obtain the tool metadata of AI tools, construct a relationship graph between tool shelf groups and AI tools, and configure resource robot units for the tool shelf groups; The shelf visibility score of the AI ​​tool is evaluated in real time, and the AI ​​tool is divided into the put-on candidate list and the take-off candidate list based on the shelf visibility score; In response to an AI tool access request from a user terminal, the list of candidate listings is displayed to the user terminal. In response to an AI tool selection request from a user terminal, the target AI tool in the AI ​​tool selection request is assigned to the corresponding resource robot unit for execution, and an event to put the AI ​​tool on the shelf or remove it is triggered based on the resource usage monitored in real time.

[0006] As an improvement to the above solution, the step of obtaining tool metadata of AI tools based on the MCP protocol, constructing a relationship graph between tool shelf groups and AI tools, and configuring resource robot units for the tool shelf groups includes: MCP client nodes submit the tool metadata of the AI ​​tools to be registered; The MCP server node parses the tool metadata, generates a tool context object, and stores the tool context object in the tool metadata database node; The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-cycle fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node; The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot node; The resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command.

[0007] As an improvement to the above solution, the tool metadata includes tool function vectors, tool runtime resource requirements, and tool runtime mode identifiers; The tool function vector represents the tool operation characteristics of the AI ​​tool, the tool operation resource requirement represents the operation resource requirement of the AI ​​tool, and the tool operation mode identifier includes the priority category information of the AI ​​tool.

[0008] As an improvement to the above scheme, the tool context object includes a unique tool identifier, a tool function vector, tool runtime resource requirements, and a tool runtime mode.

[0009] As an improvement to the above scheme, the MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node, including: Initialize the soft attribution coefficient matrix; wherein, the elements of the soft attribution coefficient matrix represent the fuzzy coefficients by which the AI ​​tool is assigned to the tool shelf group; Initialize the center vector of each tool shelf group; wherein each tool shelf group represents an AI application type; Iterate through each AI tool and each tool shelf group, and calculate the Euclidean distance between the current AI tool and the current tool shelf group; Calculate the soft attribution coefficient based on the Euclidean distance; Based on the soft affiliation coefficient, the structure of the current tool shelf group is divided, merged, or maintained, and the corresponding tool shelf group is assigned to the current AI tool; Based on the current attribution results of the AI ​​tools, dynamically update the center vector of all tool shelf groups; When the change in the center vector of all tool shelf groups is less than a preset change threshold in a first preset number of iterations, or when the number of iterations reaches a second preset number, the iteration process is terminated and the structure of all tool shelf groups is locked; wherein, the second preset number is greater than the first preset number. The AI ​​tools and the tool shelf groups are mapped to generate a relationship graph between the tool shelf groups and the AI ​​tools, and stored in the tool metadata database node.

[0010] As an improvement to the above solution, the step of dividing, merging, or maintaining the structure of the current tool shelf group based on the soft affiliation coefficient, and assigning a corresponding tool shelf group to the current AI tool, includes: When the soft affiliation coefficient is greater than a preset coefficient threshold, the current AI tool is affixed to the current tool shelf group; When the soft affiliation coefficient is less than or equal to a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is less than or equal to a preset number threshold, the current tool shelf group is divided into a first sub-tool shelf group and a second sub-tool shelf group, and the current AI tool is affixed to the first sub-tool shelf group or the second sub-tool shelf group. When the soft attribution coefficient is greater than a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is greater than a preset number threshold, the current tool shelf group is merged with other tool shelf groups with the highest similarity to obtain a combined tool shelf group, and the current AI tool is assigned to the combined tool shelf group.

[0011] As an improvement to the above solution, after locking the structure of all tool shelf groups, the method further includes: When any AI tool does not belong to any tool shelf group, calculate the distance between the tool function vector of the AI ​​tool and the center vector of the existing tool shelf group; Determine whether the minimum distance among the distances is less than a preset distance threshold; If so, assign the AI ​​tool to the tool shelf group corresponding to the minimum distance; If not, create a new tool shelf group and assign the AI ​​tool to the new tool shelf group.

[0012] As an improvement to the above scheme, the MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot nodes, including: The MCP coordinator node obtains the relationship graph between the tool shelf group and the AI ​​tools from the tool metadata database node, determines the set of AI tools corresponding to the tool shelf group, and extracts the tool operation resource requirements of each AI tool. Calculate the resource requirement vector of the tool shelf group based on the tool operation resource requirements of the AI ​​tools in the tool shelf group; Calculate the computational flow density of the tool shelf group based on the resource demand vector; Based on the computation flow density and resource demand vector of the tool shelf group, select and bind a resource robot unit from the resource robot unit pool; Based on the binding relationship between the tool shelf group and the resource robot unit, a resource allocation initialization command is generated and sent to the resource robot node.

[0013] As an improvement to the above solution, the resource robot unit includes lightweight, standard, heavy-duty, and dynamic types; among them, the dynamic resource robot node is connected to the elastic resource pool and can dynamically apply for and release resources; The step of selecting and binding a resource robot unit from the resource robot unit pool based on the computation flow density and the resource demand vector of the tool shelf group includes: When the computation flow density is less than or equal to a preset first computation flow density threshold, a lightweight resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset first computation flow density threshold and less than or equal to a preset second computation flow density threshold, a standard-level resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset second computation flow density threshold, a heavyweight resource robot unit is selected from the resource robot unit pool for binding; the second computation flow density threshold is greater than the first computation flow density threshold. When the variance of the resource demand vector is greater than a preset variance threshold, a dynamic resource robot unit is selected from the resource robot unit pool for binding.

[0014] As an improvement to the above solution, the resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command, including: The resource robot node responds to the resource allocation initialization command by creating a data structure for the corresponding resource robot unit of the tool shelf group; Read parameter data of all AI tools in the current tool shelf group from the tool meta-database node, and update its own task list and scheduling queue; The binding status and resource quota information are fed back to the tool's metadata database node in real time.

[0015] As an improvement to the above solution, the real-time evaluation of the shelf visibility score of the AI ​​tool, and the division of the AI ​​tool into a list of candidates for listing and a list of candidates for delisting based on the shelf visibility score, includes: The cloud disk server node calculates the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score. The cloud disk server node sorts the shelf visibility scores in descending order, adds the top K AI tools with shelf visibility scores greater than or equal to a preset first visibility score threshold to the list of candidates for listing, and adds AI tools with shelf visibility scores less than a preset second visibility score threshold to the list of candidates for delisting; wherein, the first visibility score threshold is greater than the second visibility score threshold.

[0016] As an improvement to the above solution, the cloud disk server node calculates an AI supermarket demand score for each AI tool based on relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm, including: By counting the total number of times the AI ​​tool was invoked by users within the current time window; The corresponding weighting coefficients are determined based on the type of AI tool; wherein, the weighting coefficients are used to distinguish the degree of influence of different types of AI tools on the AI ​​supermarket demand score; The degree of association between the AI ​​tool and the tool shelf group is calculated, and a weighted coefficient is obtained by weighting using a normalized soft attribution coefficient. The AI ​​supermarket demand score is obtained by multiplying the total number of times the AI ​​tool is called by the user, the weight coefficient, and the weighting coefficient.

[0017] As an improvement to the above solution, the cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score, including: Calculate the load status parameters based on the computational flow density of the tool shelf group to which the AI ​​tool belongs; Calculate the normalized Euclidean distance based on the tool function vector of the AI ​​tool and the center vector of the tool shelf group to which it belongs; The shelf visibility score of the AI ​​tool is calculated based on the AI ​​supermarket demand score, the load condition parameter, and the normalized Euclidean distance.

[0018] As an improvement to the above solution, in response to an AI tool selection request from a user terminal, the target AI tool in the AI ​​tool selection request is assigned to the corresponding resource robot unit for execution, and an on-demand or off-demand event for the AI ​​tool is triggered based on real-time monitored resource usage, including: In response to the AI ​​tool selection request from the user terminal, the cloud disk server node assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution; The resource robot node monitors the resource utilization rate of the target AI tool in real time; When the resource utilization rate of the target AI tool exceeds a preset first resource utilization rate threshold, the resource robot node sends a delisting signal to the cloud disk server node. In response to the delisting signal, the cloud disk server node updates the shelf visibility score of the target AI tool, marks the target AI tool as temporarily delisted, and removes it from the list of listing candidates. When the resource utilization rate of the target AI tool is less than a preset second resource utilization rate threshold, the resource robot node sends an uploading signal to the cloud disk server node; the second resource utilization rate threshold is less than the first resource utilization rate threshold. In response to the listing signal, the cloud disk server node updates the shelf visibility score of the target AI tool and reclassifies the target AI tool into the listing candidate list; The cloud disk server node counts the cumulative number of times the target AI tool has been removed from the platform. When the cumulative number of removals exceeds a third preset number, the target AI tool is added to the removal candidate list.

[0019] This invention also provides a management system for AI tools, the system including MCP-related nodes and cloud disk server nodes; wherein, The MCP-related nodes obtain the tool metadata of AI tools based on the MCP protocol, construct a relationship graph between the tool shelf group and AI tools, and configure resource robot units for the tool shelf group. The cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into the listing candidate list and the delisting candidate list based on the shelf visibility score; The cloud disk server node responds to the user terminal's AI tool access request by displaying the list of candidate listings to the user terminal. The cloud disk server node responds to the AI ​​tool selection request from the user terminal, assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution, and triggers the listing or delisting event of the AI ​​tool based on the real-time monitored resource usage.

[0020] This invention also provides an AI tool management device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the AI ​​tool management method as described in any of the above embodiments.

[0021] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute a management method for the AI ​​tool as described in any of the above embodiments.

[0022] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the management method of the AI ​​tool as described in any of the above embodiments.

[0023] Compared with existing technologies, the AI ​​tool management method, system, device, medium, and program products disclosed in this invention introduce the MCP protocol to build a standardized tool context communication mechanism, utilize resource robot units to complete intelligent binding and scheduling of resources, and perform visibility scoring of AI tools as the basis for AI tool listing and delisting. This enables automated registration of AI tools, adaptive resource allocation, and dynamic control of tool listing and delisting, solving the problems of tool registration process relying on manual classification and inability to respond in real time to resource pressure and changes in user demand. It significantly improves the real-time response, resource utilization, and management expansion capabilities of the AI ​​supermarket system, and enhances the intelligence level of tool management and the consistency of user experience. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a management method for an AI tool provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating a preferred method for managing AI tools in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the multi-loop fuzzy attribution algorithm in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the process of binding and configuring the resource robot unit in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an AI tool management system provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0027] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0028] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0029] See Figure 1 This is a flowchart illustrating a management method for AI tools provided in an embodiment of the present invention. The embodiment of the present invention provides a management method for AI tools, including steps S11 to S13: S11. Obtain the tool metadata of AI tools based on the MCP protocol, construct the relationship graph between the tool shelf group and the AI ​​tools, and configure resource robot units for the tool shelf group; S12. Evaluate the shelf visibility score of the AI ​​tool in real time, and classify the AI ​​tool into the shelf-listing candidate list and the delisting candidate list based on the shelf visibility score; S13. In response to the AI ​​tool access request from the user terminal, the list of candidate listings is displayed to the user terminal. S14. In response to the AI ​​tool selection request from the user terminal, the target AI tool in the AI ​​tool selection request is assigned to the corresponding resource robot unit for execution, and an uploading or deloading event for the AI ​​tool is triggered based on the resource usage monitored in real time.

[0030] It's important to note that MCP (Model Context Protocol) is a model context protocol designed to unify the communication protocol between large models and external data sources and tools. The main purpose of MCP is to address the current challenge of AI models being unable to fully realize their potential due to data silos. MCP enables AI applications to securely access and manipulate local and remote data, providing an interface for AI applications to connect to everything.

[0031] This proposal introduces MCP (Multi-Channel Programming) to create an AI supermarket, treating AI tools as "goods," user visits as "patronizing the supermarket," and computing resources as "robot employees." The solution includes three core processes: dynamic tool registration, resource load assessment, and adaptive tool delisting / removal. The overall interaction involves user terminals, MCP-related nodes, cloud disk server nodes, tool metadata database nodes, and resource robot nodes.

[0032] In this embodiment of the invention, the first stage is the tool dynamic registration process, in which the MCP-related nodes, together with the tool metadata database nodes and the resource robot nodes, obtain the tool metadata of the AI ​​tools based on the MCP protocol, construct the relationship graph between the tool shelf group and the AI ​​tools, and configure resource robot units for the tool shelf group. The main purpose of this process is to handle the automatic registration and initial classification of AI tools and avoid manual intervention.

[0033] The second stage is the resource load assessment process, in which the cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into the listing candidate list and the delisting candidate list based on the shelf visibility score. The main purpose of this process is to monitor resource load in real time and assess tool demand.

[0034] The third stage is the adaptive tool delisting and removal process. In response to user terminal requests, the cloud disk server node, in conjunction with the tool metadata database node and the resource robot node, displays the AI ​​supermarket and manages the delisting and removal of AI tools. Specifically, in response to a user terminal's AI tool access request, the candidate list is displayed; in response to a user terminal's AI tool selection request, the target AI tool in the selection request is assigned to the corresponding resource robot unit for execution, and delisting events are triggered based on real-time monitoring of resource usage. This process performs dynamic delisting and removal, adjusting tool availability based on real-time feedback, and using user access events as triggers to achieve an "automated unmanned supermarket."

[0035] By employing the technical means of this invention, a standardized tool context communication mechanism is constructed by introducing the MCP protocol. The resource robot unit completes the intelligent binding and scheduling of resources and performs visibility scoring of AI tools as the basis for AI tool listing and delisting. This enables automated registration of AI tools, adaptive resource allocation, and dynamic control of tool listing and delisting. It solves the problems of tool registration relying on manual classification and being unable to respond to resource pressure and changes in user needs in real time. This significantly improves the real-time response, resource utilization, and management expansion capabilities of the AI ​​supermarket system, and enhances the intelligence level of tool management and the consistency of user experience.

[0036] As a preferred embodiment, the present invention further implements the technology of the above embodiments, wherein the MCP-related nodes include MCP client nodes, MCP server nodes, and MCP coordinator nodes.

[0037] The method described in this embodiment of the invention involves collaborative interaction between multiple functional nodes, specifically including: MCP Client Node: Initiated by the AI ​​Integrated Development Environment (IDE) or AI Tool SDK used by the developer to register and call tools; MCP Server Node: Receives and parses tool metadata to generate standardized tool context objects; MCP Orchestrator Node: As the core decision-making node, it is responsible for dynamically generating shelf groups and determining tool ownership based on a multi-cycle fuzzy attribution algorithm; Resource-Robot Node: Serves as the unit that carries the resources for tool operation, performing resource binding, status monitoring, and dynamic scheduling; Tool Metadata DB Node: Used to persistently store tool context information, shelf group mapping relationships, and resource binding status.

[0038] Step S11, namely obtaining the tool metadata of AI tools based on the MCP protocol, constructing the relationship graph between the tool shelf group and the AI ​​tools, and configuring resource robot units for the tool shelf group, includes steps S111 to S115: S111, The MCP client node submits the tool metadata of the AI ​​tool to be registered; S112. The MCP server node parses the tool metadata, generates a tool context object, and stores the tool context object in the tool metadata database node; S113. The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node. S114. The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends a resource allocation initialization command to the resource robot node. S115. The resource robot node configures a resource robot unit for the tool shelf group according to the resource allocation initialization instruction.

[0039] In this embodiment of the invention, in response to the tool submission command from the MCP client node, the MCP server node receives and parses the tool metadata, generates a tool context object, and stores it in the tool metadata database node. The MCP coordinator node reads the tool context object from the tool metadata database node in a polling manner. The system executes a Multi-Loop Fuzzy Classification (MFC) algorithm. By iterating through the Euclidean distance between each newly registered tool and the tool shelf, it calculates the soft classification coefficient and performs classification merging, subclassing, and independent grouping on the newly registered tools. It dynamically generates or updates tool shelf groups G and assigns shelf group identifiers to the current new tools, updating the results to the tool metadata database node. The MCP coordinator node sends resource allocation initialization commands to the resource robot node, binds resource robot units to the updated tool shelf group G, and determines the required computing resource quota for each shelf group based on the computation flow density and resource demand vector. After creation, each robot unit reads the allocated shelf group parameter data from the tool metadata database node and updates its own computation task list.

[0040] Preferably, the tool metadata includes a tool function vector, tool runtime resource requirements, and tool runtime mode identifier; the tool context object includes a tool unique identifier, tool function vector, tool runtime resource requirements, and tool runtime mode.

[0041] The tool function vector represents the tool operation characteristics of the AI ​​tool, the tool operation resource requirement represents the operation resource requirement of the AI ​​tool, and the tool operation mode identifier includes the priority category information of the AI ​​tool.

[0042] Specifically, when a developer clicks the "Register New Tool" button in the AI ​​development platform, the MCP client node submits the tool's metadata to the MCP server node through the standard input / output interface of the MCP protocol. This tool metadata includes, but is not limited to: Tool Function Vector: ,in The representative tool's operating characteristics include the average thinking interval time per 10,000 inference calls, the maximum RAM memory usage to VRAM usage ratio, the average number of concurrent calls per million tokens, the call frequency in the past seven days, and the functional semantic similarity with existing tools. Tool runtime resource requirements: These are the limits for CPU, memory, hard disk I / O, and video memory, respectively.

[0043] Tool operation mode identifier, including processing priority category information such as "real-time processing" or "asynchronous background".

[0044] After the MCP server node parses the metadata, it generates a tool context object. (Contextual ToolObject) and forward it to the MCP coordinator node. The CTO structure fields include: Tool unique identifier ; Tool function vector Tool and resource requirements ; Tool operation mode; Tool operation status flag (initially empty); Shelf group identifier (initially empty).

[0045] Subsequently, the tool context object It is forwarded to the MCP coordinator node to execute subsequent shelf grouping and resource mapping processes.

[0046] For a preferred embodiment, see Figure 2 This is a flowchart illustrating a preferred AI tool management method in an embodiment of the present invention. Step S113, namely, the MCP coordinator node reads the tool context object from the tool metadata database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool metadata database node, including: Initialize the soft attribution coefficient matrix; wherein, the elements of the soft attribution coefficient matrix represent the fuzzy coefficients by which the AI ​​tool is assigned to the tool shelf group; Initialize the center vector of each tool shelf group; wherein each tool shelf group represents an AI application type; Iterate through each AI tool and each tool shelf group, and calculate the Euclidean distance between the current AI tool and the current tool shelf group; Calculate the soft attribution coefficient based on the Euclidean distance; Based on the soft affiliation coefficient, the structure of the current tool shelf group is divided, merged, or maintained, and the corresponding tool shelf group is assigned to the current AI tool; Based on the current attribution results of the AI ​​tools, dynamically update the center vector of all tool shelf groups; When the change in the center vector of all tool shelf groups is less than a preset change threshold in a first preset number of iterations, or when the number of iterations reaches a second preset number, the iteration process is terminated and the structure of all tool shelf groups is locked; wherein, the second preset number is greater than the first preset number. The AI ​​tools and the tool shelf groups are mapped to generate a relationship graph between the tool shelf groups and the AI ​​tools, and stored in the tool metadata database node.

[0047] Preferably, the step of dividing, merging, or maintaining the structure of the current tool shelf group according to the soft affiliation coefficient, and assigning a corresponding tool shelf group to the current AI tool, includes: When the soft affiliation coefficient is greater than a preset coefficient threshold, the current AI tool is affixed to the current tool shelf group; When the soft affiliation coefficient is less than or equal to a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is less than or equal to a preset number threshold, the current tool shelf group is divided into a first sub-tool shelf group and a second sub-tool shelf group, and the current AI tool is affixed to the first sub-tool shelf group or the second sub-tool shelf group. When the soft attribution coefficient is greater than a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is greater than a preset number threshold, the current tool shelf group is merged with other tool shelf groups with the highest similarity to obtain a combined tool shelf group, and the current AI tool is assigned to the combined tool shelf group.

[0048] Preferably, after the structure of all tool shelf groups is locked, the method further includes: When any AI tool does not belong to any tool shelf group, calculate the distance between the tool function vector of the AI ​​tool and the center vector of the existing tool shelf group; Determine whether the minimum distance among the distances is less than a preset distance threshold; If so, assign the AI ​​tool to the tool shelf group corresponding to the minimum distance; If not, create a new tool shelf group and assign the AI ​​tool to the new tool shelf group.

[0049] In this embodiment of the invention, the MCP coordinator node receives the tool context object. Then, the MCP coordinator node initiates the Multi-Loop Fuzzy Assignment (MFM) algorithm to dynamically optimize the shelf group structure and assign the most suitable shelf group to the current tool.

[0050] The input parameters of the multi-cycle fuzzy attribution algorithm include: Tool Shelf Group Collection Each group The functional vector center is This indicates the current application status of AI in the supermarket, with each shelf group representing a type of AI application.

[0051] Current tool function vector ; Fuzzy factor ; Maximum number of iterations ; Functional similarity threshold .

[0052] See Figure 3 This is a flowchart illustrating the multi-cycle fuzzy attribution algorithm in an embodiment of the present invention. The algorithm flow of the multi-cycle fuzzy attribution algorithm is as follows: (1) Initialize a soft attribution coefficient matrix Its dimension is m×n, where m is the number of current tool shelf groups and n is the total number of tools to be registered (including current tool j); The matrix In, elements Representation tools Belongs to the tool shelf group The fuzzy coefficient is initialized using the element. .

[0053] (2) Based on the current application of AI supermarket shelf placement, initialize the center vector of each tool shelf group. Each tool shelf group This represents a type of AI application, such as image processing, speech recognition, and document summarization. center vector That is, each group center is composed of a function vector. With resource demand vector Combined composition; If the current shelf group Number of tools (e.g., η=3), then use The tools already exist in China Weighted average generation .

[0054] (3) Enter the main loop iteration process, with the number of iterations not exceeding the set value MaxIterations (e.g., 50 times). In each iteration, each tool is processed sequentially. (j=1 to n) Perform the following processing: a. Execute the inner group traversal loop, for each tool shelf group (i=1 to m) Calculation tools With shelving groups Euclidean distance:

[0055] b. Based on Euclidean distance Execute multi-branch decision logic and calculate soft attribution coefficient. :

[0056] If the soft attribution coefficient (e.g., 0.3), then the tool Belong to shelf group ; If the soft attribution coefficient And the current shelf group quantity (e.g., 4) then create a subclass shelf group. Divide the current shelf group equally Tools, put tools Randomly add it to one of the subclass shelf groups and update the center vector. Recalculate the soft attribution coefficients of each tool after splitting. ; If the soft attribution coefficient And the current shelf group quantity Then find the items in all current shelf groups that are related to the shelf group. The other group with the highest similarity (Right now ); group the shelves With shelving groups Merged into a new shelving group Put the tools Join the new shelf group Update its center vector Update the soft attribution matrix .

[0057] (4) Based on the attribution results in (3) above, dynamically update the shelf group center vector. :

[0058] (5) If the change in the center vector of all tool shelf groups is in three consecutive iterations If the value is 0.01, then the current iteration process will be terminated and the current shelf group structure will be locked.

[0059] (6) If tools If the device is not assigned to any group in the current iteration, the following multi-branch logic is executed: a. Determine the current tool Functional vector Is the minimum distance from the center vector of the existing shelf group less than... (e.g., 0.5), if true, then the tool Map to the nearest shelf group Adjust its group center vector If not, then use tools. Create a new shelf group at the center Set its center vector as .

[0060] (7) Set the current tool of With shelving groups or Perform mapping, construct a shelf group-tool relationship graph, and write this mapping relationship into the tool metadata database node; if the tool Assigned to multiple shelving groups (i.e. ): like (e.g., 0.1), then it belongs to Higher groups; like Approaching (i.e., difference) If ), it means that the tool can be referenced by multiple shelves.

[0061] Using the technical means of this invention, an innovative method for dynamic registration and shelf grouping of AI tools based on a multi-cycle fuzzy attribution algorithm is proposed. The MCP coordinator node reads the tool context object of the tool metadata database node in a round-robin manner. The method employs a multi-loop fuzzy attribution algorithm, iteratively traversing the Euclidean distance between each newly registered tool and the tool shelf to calculate its soft attribution coefficient. It then performs attribution merging, subclass segmentation, and independent grouping on the newly registered tools, dynamically generating or updating tool shelf groups and assigning shelf group identifiers to the current new tools. The results are then updated to the tool metadata database nodes. Compared to existing technologies, this method solves the problems of reliance on manual classification in the tool registration process, fixed shelf grouping, and inability to adapt to the evolution of tool functions. It improves the automation level of tool registration, the flexibility of shelf structure, and the proactive nature of resource scheduling.

[0062] In a preferred embodiment, step S114 involves the MCP coordinator node binding resource robot units to the updated tool shelf group based on the relationship graph, and sending a resource allocation initialization command to the resource robot node, including: The MCP coordinator node obtains the relationship graph between the tool shelf group and the AI ​​tools from the tool metadata database node, determines the set of AI tools corresponding to the tool shelf group, and extracts the tool operation resource requirements of each AI tool. Calculate the resource requirement vector of the tool shelf group based on the tool operation resource requirements of the AI ​​tools in the tool shelf group; Calculate the computational flow density of the tool shelf group based on the resource demand vector; Based on the computation flow density and resource demand vector of the tool shelf group, select and bind a resource robot unit from the resource robot unit pool; Based on the binding relationship between the tool shelf group and the resource robot unit, a resource allocation initialization command is generated and sent to the resource robot node.

[0063] Preferably, the resource robot units include lightweight, standard, heavy-duty, and dynamic types; wherein, the dynamic resource robot nodes are connected to the elastic resource pool and can dynamically apply for and release resources; The step of selecting and binding a resource robot unit from the resource robot unit pool based on the computation flow density and the resource demand vector of the tool shelf group includes: When the computation flow density is less than or equal to a preset first computation flow density threshold, a lightweight resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset first computation flow density threshold and less than or equal to a preset second computation flow density threshold, a standard-level resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset second computation flow density threshold, a heavyweight resource robot unit is selected from the resource robot unit pool for binding; the second computation flow density threshold is greater than the first computation flow density threshold. When the variance of the resource demand vector is greater than a preset variance threshold, a dynamic resource robot unit is selected from the resource robot unit pool for binding.

[0064] In a preferred embodiment, step 115, namely, the resource robot node configuring resource robot units for the tool shelf group according to the resource allocation initialization instruction, includes: The resource robot node responds to the resource allocation initialization command by creating a data structure for the corresponding resource robot unit of the tool shelf group; Read parameter data of all AI tools in the current tool shelf group from the tool meta-database node, and update its own task list and scheduling queue; The binding status and resource quota information are fed back to the tool's metadata database node in real time.

[0065] In an embodiment of the present invention, see Figure 4 This is a flowchart illustrating the process of binding and configuring resource robot units in an embodiment of the present invention. (1) The MCP coordinator node obtains the shelf group from the tool metadata node. toolset And extract the function vectors of each tool. With resource demand vector ; Function vector This includes dimensions such as average thinking interval time per 10,000 inference calls, maximum RAM memory usage to VRAM usage ratio, average concurrent calls per million tokens, call frequency in the past seven days, and functional semantic similarity with existing tools. Resource demand vector This represents the average resource consumption during tool runtime.

[0066] (2) Based on shelf groups Resource requirement vector set of tools Calculate the group resource demand vector ; If the shelf group For newly generated groups (such as ),but (Resource vector of new tool j); If the shelf group is a new structure after splitting / merging, then the calculation is based on the tool distribution during the splitting / merging operation. ; Determine the "calculation flow density" : This is used for subsequent resource allocation decisions.

[0067] (3) According to the shelving group Calculated flow density With resource demand vector The MCP coordinator node selects and binds a resource robot unit from the resource robot unit pool, wherein the resource robot unit is in the form of RRU (Remote Radio Unit).

[0068] The following are examples of RRU type elements contained in the resource robot unit pool: RRU-Lite (Lightweight): CPU=2 cores, memory=4GB, IO=100ms; RRU-Standard (Standard Level): CPU=4 cores, memory=16GB, I / O=200ms; RRU-Heavy: CPU=8 cores, memory=64GB, I / O=500ms; RRU-Dynamic: Connects to the elastic resource pool and can dynamically request / release resources.

[0069] The binding strategy between shelf groups and resource robot units is as follows: like If the tool's concurrency requirement is less than 5, then bind the lightweight RRU-Lite; like If the tool's concurrency requirement is less than 20, then it is bound to the standard RRU-Standard level. like If the tool's concurrent requirements are greater than 20, then bind it to the heavy-duty RRU-Heavy class; like If the variance value is greater than 20, then bind the dynamically allocated RRU-Dynamic.

[0070] (4) The MCP coordinator node constructs initialization instructions and sends them to the resource robot node through the MCP protocol remote communication mechanism (SSE / HTTP combination).

[0071] (5) After receiving the initialization command, the resource robot node performs the following operations: a. Create a resource robot cell in its local operating environment ( The data structure contains the following fields: Bind shelf group ID ( ); Current resource quota ( ); Tool Task List (Pending) (list) Resource damping strategies (thresholds, inspection cycles); Virtual resource topology location information with other resource robot units (used for subsequent resource scheduling and tool removal / removal determination); like It is bound to a new shelf group (e.g.) If so, a new RRU unit will be dynamically requested from the elastic resource pool, and the initial resource quota will be set to [value]. Reserve some resources; after binding is complete, send a confirmation message to the MCP coordinator node.

[0072] b. Resource Robot Unit Once created, the current shelf group will be read from the tool metadata node. Update its own task list and scheduling queue by analyzing the parameter data of all tools in the system. If a certain tool CPU usage exceeding [a certain threshold] in two consecutive scheduling cycles If so, mark it as a "resource damping candidate"; If a certain tool If a item is marked as a "resource damping candidate", it will be removed from the current shelf group. And try to assign it to other groups.

[0073] (6) The resource robot node will feed back the binding status and resource quota information to the tool meta database node in real time.

[0074] This invention innovatively proposes a dynamic resource binding and computation flow density evaluation method based on Resource Robot Units (RRUs). The MCP coordinator node sends resource allocation initialization commands to the resource robot nodes. Based on the computation flow density and resource demand vector of shelf groups, the method intelligently selects the resource robot unit type and dynamically initializes its resource quota, task queue, and scheduling strategy, establishing a binding relationship between resources and tool shelves. After creation, each robot unit reads the allocated shelf group parameter data from the tool metadata node and updates its own computation task list. Compared to existing technologies, this method solves the problems of static resource allocation, inability to adapt to changes in tool load, and low resource utilization, improving resource scheduling efficiency, tool operation stability, and system scalability.

[0075] As a preferred embodiment, the present invention further implements the technology of any of the above embodiments. Step S12, namely, the real-time evaluation of the shelf visibility score of the AI ​​tool, and the division of the AI ​​tool into the shelf-listing candidate list and the delisting candidate list based on the shelf visibility score, includes steps S121 to S123: S121. The cloud disk server node calculates the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm. S122. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score. S123. The cloud disk server node sorts the shelf visibility scores in descending order, adds the top K AI tools with shelf visibility scores greater than or equal to a preset first visibility score threshold to the list of candidates for listing, and adds AI tools with shelf visibility scores less than a preset second visibility score threshold to the list of candidates for delisting; wherein, the first visibility score threshold is greater than the second visibility score threshold.

[0076] In this embodiment of the invention, the resource robot node periodically collects the "computation flow density" data of all tool shelf groups and reports it to the cloud disk server node. The resource robot node collects data (e.g., the current task queue length) every 5 seconds. Average response delay Resource utilization rate Calculate the "Computational Flow Density" (The calculation method is described in the previous stage.) If If the threshold (e.g., 0.8) is reached, an emergency report will be triggered; otherwise, a regular report will be submitted.

[0077] After receiving the data reported by the resource robot node, the cloud disk server node evaluates the priority of AI tool demand in real time based on the time window and consensus binding algorithm, generates a "shelf visibility score", and generates a "list of candidate items for listing and delisting" based on this score and presents it to the user terminal.

[0078] In a preferred implementation, step S121, namely, the cloud disk server node calculating the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm, includes: By counting the total number of times the AI ​​tool was invoked by users within the current time window; The corresponding weighting coefficients are determined based on the type of AI tool; wherein, the weighting coefficients are used to distinguish the degree of influence of different types of AI tools on the AI ​​supermarket demand score; The degree of association between the AI ​​tool and the tool shelf group is calculated, and a weighted coefficient is obtained by weighting using a normalized soft attribution coefficient. The AI ​​supermarket demand score is obtained by multiplying the total number of times the AI ​​tool is called by the user, the weight coefficient, and the weighting coefficient.

[0079] In this embodiment of the invention, after receiving the tool running status data reported by the resource robot node, the cloud disk server node executes a resource-function-user behavior joint evaluation process, constructs an Extended Consensus Bundling Algorithm (ECBBA), and generates a tool shelf visibility score by combining it with a time window mechanism. It is used for dynamic product listing and delisting decisions and for user-accessible tool control.

[0080] The core input parameters used in this step include: Shelf group Read from the tool metadata node; Tool Function Vector From step one Extract from; Tool resource requirement vector : Defined by the resource binding process in the registration phase of step one; Current shelf group resource load Reports are submitted in real time from the resource robot unit; Tool call frequency in user access logs : Indicates the past The number of times all tools within the window have been invoked; Tool scheduling status flags Record whether the tool is in the resource damping candidate list in step three; Soft attribution coefficient of tools : The result of the fuzzy classification in the previous stage; Shelf group center vector : The result generated in the previous stage.

[0081] The processing steps described in this step are as follows: Cloud disk server node defines time window The default time is 10 minutes before and after the current time, used to collect user behavior logs and tool scheduling data; Based on the data collected within the window, the cloud disk server node is for each tool. Calculate AI supermarket demand score It is a comprehensive metric that measures the frequency with which a tool is clicked or invoked by users and its load status. Among them, the demand score of AI supermarket The construction method first involves statistically analyzing the current time window. Internal tools Total number of times called by users Then, determine the corresponding weighting coefficients based on the tool type. This is used to differentiate the degree of influence of different types of tools on the score results; then, the correlation between the tool and the shelf group is calculated, using a normalized soft attribution coefficient. The weighted average is then applied; finally, the three elements are multiplied together to obtain the AI ​​supermarket demand score. :

[0082] in, For tools The total number of times the message was invoked by the user within the current time window; For window span (e.g., 20 minutes); The tool type weight is set (e.g., 1.2 for image processing and 0.8 for asynchronous background tools). For tools Belonging to shelf group The soft attribution coefficient; The largest attribution coefficient among all tools is used for normalization.

[0083] In a preferred embodiment, step S122, namely, the cloud disk server node generating a shelf visibility score for each AI tool based on the AI ​​supermarket demand score, includes: Calculate the load status parameters based on the computational flow density of the tool shelf group to which the AI ​​tool belongs; Calculate the normalized Euclidean distance based on the tool function vector of the AI ​​tool and the center vector of the tool shelf group to which it belongs; The shelf visibility score of the AI ​​tool is calculated based on the AI ​​supermarket demand score, the load condition parameter, and the normalized Euclidean distance.

[0084] In this embodiment of the invention, the AI ​​supermarket demand score is used as the basis. Generate shelf visibility score for each tool. It is a tool for AI supermarket demand score Compared with the current resource flow density The weighting function; Among them, shelf visibility score The construction method is first based on the AI ​​supermarket demand score. As input, combined with the current resource flow density of the shelf group ,pass This item takes load conditions into account (the more abundant the resources, the greater the contribution of this item); it also calculates the normalized Euclidean distance between the tool function vector and the center of the shelf group. Finally, the shelf visibility score is obtained by linearly weighting the two types of indicators using a balancing factor δ (default 0.6). :

[0085] in, This is a balancing factor used to adjust the weight of resource load and user behavior; its initial value is 0.6. The Euclidean distance between the tool function vector and the center of the shelf group; Let be the magnitude of the center vector of the shelf group; To indicate current resource availability, the closer the availability is to 0, the more idle the resource is. The project introduces non-linearity to prevent high-functionality but high-load tools from being mistakenly listed.

[0086] Cloud disk server nodes are based on generated shelf visibility scores The system executes multi-level sorting and constraint logic to generate lists of candidates for shelf placement and removal. All tools are ranked according to shelf visibility score. Sort in descending order; select shelf visibility score The former Several tools were selected as "candidates for listing"; among them,

[0087] If the tool has two consecutive rounds of shelf visibility score If the value is 0.35, it will be placed in the "removal candidate pool".

[0088] This invention innovatively proposes a dynamic evaluation method for the listing and delisting of AI tools based on the Extended Consensus Bundling Algorithm (ECBBA) and a time window mechanism. After receiving data reported by the resource robot node, the cloud disk server node constructs an AI supermarket demand score and shelf visibility score based on the time window and consensus bundling algorithm, combining user access behavior, tool function matching degree, and current resource load status. A candidate list for listing and delisting is then generated based on multi-level sorting and constraint logic. Compared to existing technologies, this method solves the problems of tool listing and delisting relying on manual judgment and being unable to respond in real time to resource pressure and changes in user demand, thus improving the intelligence level of tool management, resource utilization efficiency, and user experience consistency.

[0089] As a preferred embodiment, the present invention further implements the technology of any of the above embodiments. Step S14, namely, responding to the AI ​​tool selection request from the user terminal, assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution, and triggers an uploading or deloading event for the AI ​​tool based on the real-time monitored resource usage, including steps S141 to S147: S141. The cloud disk server node responds to the AI ​​tool selection request from the user terminal and assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution. S142. The resource robot node monitors the resource occupancy rate of the target AI tool in real time; S143. When the resource occupancy rate of the target AI tool is greater than the preset first resource occupancy rate threshold, the resource robot node sends a delisting signal to the cloud disk server node. S144. In response to the delisting signal, the cloud disk server node updates the shelf visibility score of the target AI tool, marks the target AI tool as temporarily delisted, and removes it from the list of listing candidates. S145. When the resource utilization rate of the target AI tool is less than the preset second resource utilization rate threshold, the resource robot node sends an uploading signal to the cloud disk server node; the second resource utilization rate threshold is less than the first resource utilization rate threshold. S146. In response to the listing signal, the cloud disk server node updates the shelf visibility score of the target AI tool and reclassifies the target AI tool into the listing candidate list. S147. The cloud disk server node counts the cumulative number of times the target AI tool has been removed from the platform. When the cumulative number of removals exceeds a third preset number, the target AI tool is added to the removal candidate list.

[0090] In this embodiment of the invention, the user terminal sends an access request to the server node, and the cloud disk server node determines the access level based on the current shelf visibility score. Generate a visible list of tools, displaying only shelf visibility scores. tools ( Configure items for operations personnel, such as 0.5).

[0091] When a user selects a tool and clicks on it, the cloud disk server node assigns a task to the corresponding resource robot unit. When the resource robot unit executes the task, if the resource utilization rate... If the value is 0.9, a "removal signal" is sent to the cloud disk server node, specifying the tool ID to be removed. Upon receiving the signal, the cloud disk server node immediately updates the shelf visibility score. The tool will be marked as "temporarily removed" and will be relisted once the resource usage rate returns to 0.5. If the resource usage rate exceeds the limit three times consecutively, the tool will be permanently removed.

[0092] By employing the technical means of this invention, a standardized tool context communication mechanism is constructed by introducing the MCP protocol. Combined with a multi-cycle fuzzy affiliation algorithm, intelligent grouping of tools and dynamic optimization of shelf groups are achieved. Resource robot units are used to complete intelligent binding and scheduling of resources. Based on the Extended Consensus Bundling Algorithm (ECBBA), user behavior and resource load status are integrated to perform tool visibility scoring. This enables automated registration of AI tools, adaptive allocation of resources, and dynamic control of tool uploading and unloading, significantly improving the real-time response, resource utilization, and management scalability of the AI ​​supermarket system.

[0093] It should be noted that the preset thresholds in all embodiments of the present invention can be set according to actual application conditions or empirical values, and are not specifically limited here.

[0094] See Figure 5 This is a schematic diagram of the structure of an AI tool management system provided in an embodiment of the present invention. The system includes MCP-related nodes and cloud disk server nodes; wherein... The MCP-related nodes obtain the tool metadata of AI tools based on the MCP protocol, construct a relationship graph between the tool shelf group and AI tools, and configure resource robot units for the tool shelf group. The cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into the listing candidate list and the delisting candidate list based on the shelf visibility score; The cloud disk server node responds to the user terminal's AI tool access request by displaying the list of candidate listings to the user terminal. The cloud disk server node responds to the AI ​​tool selection request from the user terminal, assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution, and triggers the listing or delisting event of the AI ​​tool based on the real-time monitored resource usage.

[0095] By employing the technical means of this invention, a standardized tool context communication mechanism is constructed by introducing the MCP protocol. The resource robot unit completes the intelligent binding and scheduling of resources and performs visibility scoring of AI tools as the basis for AI tool listing and delisting. This enables automated registration of AI tools, adaptive resource allocation, and dynamic control of tool listing and delisting. It solves the problems of tool registration relying on manual classification and being unable to respond to resource pressure and changes in user needs in real time. This significantly improves the real-time response, resource utilization, and management expansion capabilities of the AI ​​supermarket system, and enhances the intelligence level of tool management and the consistency of user experience.

[0096] In a preferred embodiment, the MCP-related nodes include MCP client nodes, MCP server nodes, and MCP coordinator nodes; the system also includes tool metadata database nodes and resource robot nodes. The MCP-related nodes then obtain the tool metadata of the AI ​​tools based on the MCP protocol, construct a relationship graph between the tool shelf group and the AI ​​tools, and configure resource robot units for the tool shelf group, specifically: The MCP client node submits the tool metadata of the AI ​​tool to be registered; The MCP server node parses the tool metadata, generates a tool context object, and stores the tool context object in the tool metadata database node; The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node; The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot node; The resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command.

[0097] In a preferred embodiment, the tool metadata includes a tool function vector, tool runtime resource requirements, and tool runtime mode identifier; Wherein, the tool function vector represents the tool operation characteristics of the AI ​​tool, the tool operation resource requirement represents the operation resource requirement of the AI ​​tool, and the tool operation mode identifier includes the priority category information of the AI ​​tool; The tool context object includes a unique tool identifier, a tool function vector, tool runtime resource requirements, and a tool runtime mode.

[0098] In a preferred embodiment, the MCP coordinator node is specifically used for: Initialize the soft attribution coefficient matrix; wherein, the elements of the soft attribution coefficient matrix represent the fuzzy coefficients by which the AI ​​tool is assigned to the tool shelf group; Initialize the center vector of each tool shelf group; wherein each tool shelf group represents an AI application type; Iterate through each AI tool and each tool shelf group, and calculate the Euclidean distance between the current AI tool and the current tool shelf group; Calculate the soft attribution coefficient based on the Euclidean distance; Based on the soft affiliation coefficient, the structure of the current tool shelf group is divided, merged, or maintained, and the corresponding tool shelf group is assigned to the current AI tool; Based on the current attribution results of the AI ​​tools, dynamically update the center vector of all tool shelf groups; When the change in the center vector of all tool shelf groups is less than a preset change threshold in a first preset number of iterations, or when the number of iterations reaches a second preset number, the iteration process is terminated and the structure of all tool shelf groups is locked; wherein, the second preset number is greater than the first preset number. The AI ​​tools and the tool shelf groups are mapped to generate a relationship graph between the tool shelf groups and the AI ​​tools, and stored in the tool metadata database node; The step of dividing, merging, or maintaining the structure of the current tool shelf group based on the soft affiliation coefficient, and assigning a corresponding tool shelf group to the current AI tool, includes: When the soft affiliation coefficient is greater than a preset coefficient threshold, the current AI tool is affixed to the current tool shelf group; When the soft affiliation coefficient is less than or equal to a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is less than or equal to a preset number threshold, the current tool shelf group is divided into a first sub-tool shelf group and a second sub-tool shelf group, and the current AI tool is affixed to the first sub-tool shelf group or the second sub-tool shelf group. When the soft attribution coefficient is greater than a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is greater than a preset number threshold, the current tool shelf group is merged with other tool shelf groups with the highest similarity to obtain a combined tool shelf group, and the current AI tool is assigned to the combined tool shelf group.

[0099] In a preferred embodiment, the MCP coordinator node is further configured to: Obtain the relationship graph between the tool shelf group and AI tools from the tool meta-database node, determine the set of AI tools corresponding to the tool shelf group, and extract the tool operation resource requirements of each AI tool; Calculate the resource requirement vector of the tool shelf group based on the tool operation resource requirements of the AI ​​tools in the tool shelf group; Calculate the computational flow density of the tool shelf group based on the resource demand vector; Based on the computation flow density and resource demand vector of the tool shelf group, select and bind a resource robot unit from the resource robot unit pool; Based on the binding relationship between the tool shelf group and the resource robot unit, a resource allocation initialization command is generated and sent to the resource robot node; The resource robot units include lightweight, standard, heavy-duty, and dynamic types; the dynamic resource robot nodes are connected to the elastic resource pool and can dynamically apply for and release resources. The step of selecting and binding a resource robot unit from the resource robot unit pool based on the computation flow density and the resource demand vector of the tool shelf group includes: When the computation flow density is less than or equal to a preset first computation flow density threshold, a lightweight resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset first computation flow density threshold and less than or equal to a preset second computation flow density threshold, a standard-level resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset second computation flow density threshold, a heavyweight resource robot unit is selected from the resource robot unit pool for binding; the second computation flow density threshold is greater than the first computation flow density threshold. When the variance of the resource demand vector is greater than a preset variance threshold, a dynamic resource robot unit is selected from the resource robot unit pool for binding.

[0100] In a preferred embodiment, the resource robot node is specifically used for: In response to the resource allocation initialization command, a data structure for the corresponding resource robot unit is created for the tool shelf group; Read parameter data of all AI tools in the current tool shelf group from the tool meta-database node, and update its own task list and scheduling queue; The binding status and resource quota information are fed back to the tool's metadata database node in real time.

[0101] In a preferred implementation, the cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into a listing candidate list and a delisting candidate list based on the shelf visibility score, specifically as follows: The cloud disk server node calculates the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score. The cloud disk server node sorts the shelf visibility scores in descending order, adds the top K AI tools with shelf visibility scores greater than or equal to a preset first visibility score threshold to the list of candidates for listing, and adds AI tools with shelf visibility scores less than a preset second visibility score threshold to the list of candidates for delisting; wherein, the first visibility score threshold is greater than the second visibility score threshold.

[0102] In a preferred embodiment, the cloud disk server node calculates an AI supermarket demand score for each AI tool based on relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm, including: By counting the total number of times the AI ​​tool was invoked by users within the current time window; The corresponding weighting coefficients are determined based on the type of AI tool; wherein, the weighting coefficients are used to distinguish the degree of influence of different types of AI tools on the AI ​​supermarket demand score; The degree of association between the AI ​​tool and the tool shelf group is calculated, and a weighted coefficient is obtained by weighting using a normalized soft attribution coefficient. The AI ​​supermarket demand score is obtained by multiplying the total number of times the AI ​​tool is called by the user, the weight coefficient, and the weighting coefficient. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score, including: Calculate the load status parameters based on the computational flow density of the tool shelf group to which the AI ​​tool belongs; Calculate the normalized Euclidean distance based on the tool function vector of the AI ​​tool and the center vector of the tool shelf group to which it belongs; The shelf visibility score of the AI ​​tool is calculated based on the AI ​​supermarket demand score, the load condition parameter, and the normalized Euclidean distance.

[0103] In a preferred embodiment, the cloud disk server node responds to the user terminal's AI tool selection request by assigning the target AI tool in the request to the corresponding resource robot unit for execution, and triggers an uploading or deloading event for the AI ​​tool based on real-time monitored resource usage, including: In response to the AI ​​tool selection request from the user terminal, the cloud disk server node assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution; The resource robot node monitors the resource utilization rate of the target AI tool in real time; When the resource utilization rate of the target AI tool exceeds a preset first resource utilization rate threshold, the resource robot node sends a delisting signal to the cloud disk server node. In response to the delisting signal, the cloud disk server node updates the shelf visibility score of the target AI tool, marks the target AI tool as temporarily delisted, and removes it from the list of listing candidates. When the resource utilization rate of the target AI tool is less than a preset second resource utilization rate threshold, the resource robot node sends an uploading signal to the cloud disk server node; the second resource utilization rate threshold is less than the first resource utilization rate threshold. In response to the listing signal, the cloud disk server node updates the shelf visibility score of the target AI tool and reclassifies the target AI tool into the listing candidate list; The cloud disk server node counts the cumulative number of times the target AI tool has been removed from the platform. When the cumulative number of removals exceeds a third preset number, the target AI tool is added to the removal candidate list.

[0104] It should be noted that the AI ​​tool management system provided in this embodiment of the invention is used to execute all process steps of the AI ​​tool management method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0105] This invention also provides an AI tool management device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the AI ​​tool management method as described in any of the above embodiments.

[0106] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute a management method for AI tools as described in any of the above embodiments.

[0107] This invention also provides a computer program product, which includes a computer program or computer instructions. When the computer program or computer instructions are executed by a processor, they implement the management method of the AI ​​tool as described in any of the above embodiments.

[0108] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0109] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A management method for AI tools, characterized in that, include: Based on the MCP protocol, obtain the tool metadata of AI tools, construct a relationship graph between tool shelf groups and AI tools, and configure resource robot units for the tool shelf groups; The shelf visibility score of the AI ​​tool is evaluated in real time, and the AI ​​tool is divided into the put-on candidate list and the take-off candidate list based on the shelf visibility score; In response to an AI tool access request from a user terminal, the list of candidate listings is displayed to the user terminal. In response to an AI tool selection request from a user terminal, the target AI tool in the AI ​​tool selection request is assigned to the corresponding resource robot unit for execution, and an event to put the AI ​​tool on the shelf or remove it is triggered based on the resource usage monitored in real time.

2. The management method for AI tools as described in claim 1, characterized in that, The process of acquiring tool metadata for AI tools based on the MCP protocol, constructing a relationship graph between tool shelf groups and AI tools, and configuring resource robot units for the tool shelf groups includes: MCP client nodes submit the tool metadata of the AI ​​tools to be registered; The MCP server node parses the tool metadata, generates a tool context object, and stores the tool context object in the tool metadata database node; The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns the corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node; The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot node; The resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command.

3. The management method for AI tools as described in claim 2, characterized in that, The tool metadata includes tool function vectors, tool runtime resource requirements, and tool runtime mode identifiers; The tool function vector represents the tool operation characteristics of the AI ​​tool, the tool operation resource requirement represents the operation resource requirement of the AI ​​tool, and the tool operation mode identifier includes the priority category information of the AI ​​tool.

4. The management method for AI tools as described in claim 3, characterized in that, The tool context object includes a unique tool identifier, a tool function vector, tool runtime resource requirements, and a tool runtime mode.

5. The management method for AI tools as described in claim 2, characterized in that, The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node, including: Initialize the soft attribution coefficient matrix; wherein, the elements of the soft attribution coefficient matrix represent the fuzzy coefficients by which the AI ​​tool is assigned to the tool shelf group; Initialize the center vector of each tool shelf group; wherein each tool shelf group represents an AI application type; Iterate through each AI tool and each tool shelf group, and calculate the Euclidean distance between the current AI tool and the current tool shelf group; Calculate the soft attribution coefficient based on the Euclidean distance; Based on the soft affiliation coefficient, the structure of the current tool shelf group is divided, merged, or maintained, and the corresponding tool shelf group is assigned to the current AI tool; Based on the current attribution results of the AI ​​tools, dynamically update the center vector of all tool shelf groups; When the change in the center vector of all tool shelf groups is less than a preset change threshold in a first preset number of iterations, or when the number of iterations reaches a second preset number, the iteration process is terminated and the structure of all tool shelf groups is locked; wherein, the second preset number is greater than the first preset number. The AI ​​tools and the tool shelf groups are mapped to generate a relationship graph between the tool shelf groups and the AI ​​tools, and stored in the tool metadata database node.

6. The management method for AI tools as described in claim 5, characterized in that, The step of dividing, merging, or maintaining the structure of the current tool shelf group based on the soft affiliation coefficient, and assigning a corresponding tool shelf group to the current AI tool, includes: When the soft affiliation coefficient is greater than a preset coefficient threshold, the current AI tool is affixed to the current tool shelf group; When the soft affiliation coefficient is less than or equal to a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is less than or equal to a preset number threshold, the current tool shelf group is divided into a first sub-tool shelf group and a second sub-tool shelf group, and the current AI tool is affixed to the first sub-tool shelf group or the second sub-tool shelf group. When the soft attribution coefficient is greater than a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is greater than a preset number threshold, the current tool shelf group is merged with other tool shelf groups with the highest similarity to obtain a combined tool shelf group, and the current AI tool is assigned to the combined tool shelf group.

7. The management method for AI tools as described in claim 5, characterized in that, After the structure of all tool shelf groups has been locked, the method further includes: When any AI tool does not belong to any tool shelf group, calculate the distance between the tool function vector of the AI ​​tool and the center vector of the existing tool shelf group; Determine whether the minimum distance among the distances is less than a preset distance threshold; If so, assign the AI ​​tool to the tool shelf group corresponding to the minimum distance; If not, create a new tool shelf group and assign the AI ​​tool to the new tool shelf group.

8. The management method for AI tools as described in claim 2, characterized in that, The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot nodes, including: The MCP coordinator node obtains the relationship graph between the tool shelf group and the AI ​​tools from the tool metadata database node, determines the set of AI tools corresponding to the tool shelf group, and extracts the tool operation resource requirements of each AI tool. Calculate the resource requirement vector of the tool shelf group based on the tool operation resource requirements of the AI ​​tools in the tool shelf group; Calculate the computational flow density of the tool shelf group based on the resource demand vector; Based on the computation flow density and resource demand vector of the tool shelf group, select and bind a resource robot unit from the resource robot unit pool; Based on the binding relationship between the tool shelf group and the resource robot unit, a resource allocation initialization command is generated and sent to the resource robot node.

9. The management method for AI tools as described in claim 8, characterized in that, The resource robot units include lightweight, standard, heavy-duty, and dynamic types; among them, dynamic resource robot nodes are connected to the elastic resource pool and can dynamically apply for and release resources. The step of selecting and binding a resource robot unit from the resource robot unit pool based on the computation flow density and the resource demand vector of the tool shelf group includes: When the computation flow density is less than or equal to a preset first computation flow density threshold, a lightweight resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset first computation flow density threshold and less than or equal to a preset second computation flow density threshold, a standard-level resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset second computation flow density threshold, a heavyweight resource robot unit is selected from the resource robot unit pool for binding; the second computation flow density threshold is greater than the first computation flow density threshold. When the variance of the resource demand vector is greater than a preset variance threshold, a dynamic resource robot unit is selected from the resource robot unit pool for binding.

10. The management method for AI tools as described in claim 2, characterized in that, The resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command, including: The resource robot node responds to the resource allocation initialization command by creating a data structure for the corresponding resource robot unit of the tool shelf group; Read parameter data of all AI tools in the current tool shelf group from the tool metadata database node, and update its own task list and scheduling queue; The binding status and resource quota information are fed back to the tool's metadata database node in real time.

11. The management method for AI tools as described in claim 1, characterized in that, The real-time evaluation of the shelf visibility score of the AI ​​tool, and the classification of the AI ​​tool into the put-on candidate list and the take-off candidate list based on the shelf visibility score, includes: The cloud disk server node calculates the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score. The cloud disk server node sorts the shelf visibility scores in descending order, adds the top K AI tools with shelf visibility scores greater than or equal to a preset first visibility score threshold to the list of candidates for listing, and adds AI tools with shelf visibility scores less than a preset second visibility score threshold to the list of candidates for delisting; wherein, the first visibility score threshold is greater than the second visibility score threshold.

12. The management method for AI tools as described in claim 11, characterized in that, The cloud disk server node calculates an AI supermarket demand score for each AI tool based on relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm, including: By counting the total number of times the AI ​​tool was invoked by users within the current time window; The corresponding weighting coefficients are determined based on the type of AI tool; wherein, the weighting coefficients are used to distinguish the degree of influence of different types of AI tools on the AI ​​supermarket demand score; The degree of association between the AI ​​tool and the tool shelf group is calculated, and a weighted coefficient is obtained by weighting using a normalized soft attribution coefficient. The AI ​​supermarket demand score is obtained by multiplying the total number of times the AI ​​tool is called by the user, the weight coefficient, and the weighting coefficient.

13. The management method for AI tools as described in claim 11, characterized in that, The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score, including: Calculate the load status parameters based on the computational flow density of the tool shelf group to which the AI ​​tool belongs; Calculate the normalized Euclidean distance based on the tool function vector of the AI ​​tool and the center vector of the tool shelf group to which it belongs; The shelf visibility score of the AI ​​tool is calculated based on the AI ​​supermarket demand score, the load condition parameter, and the normalized Euclidean distance.

14. The management method for AI tools as described in claim 2, characterized in that, The process of responding to an AI tool selection request from a user terminal involves assigning the target AI tool in the request to the corresponding resource robot unit for execution, and triggering an add-on or remove-off event for the AI ​​tool based on real-time monitored resource usage, including: In response to the AI ​​tool selection request from the user terminal, the cloud disk server node assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution; The resource robot node monitors the resource utilization rate of the target AI tool in real time; When the resource utilization rate of the target AI tool exceeds a preset first resource utilization rate threshold, the resource robot node sends a delisting signal to the cloud disk server node. In response to the delisting signal, the cloud disk server node updates the shelf visibility score of the target AI tool, marks the target AI tool as temporarily delisted, and removes it from the list of listing candidates. When the resource utilization rate of the target AI tool is less than a preset second resource utilization rate threshold, the resource robot node sends an uploading signal to the cloud disk server node; the second resource utilization rate threshold is less than the first resource utilization rate threshold. In response to the listing signal, the cloud disk server node updates the shelf visibility score of the target AI tool and reclassifies the target AI tool into the listing candidate list; The cloud disk server node counts the cumulative number of times the target AI tool has been removed from the platform. When the cumulative number of removals exceeds a third preset number, the target AI tool is added to the removal candidate list.

15. A management system for an AI tool, characterized in that, The system includes MCP-related nodes and cloud disk server nodes; wherein... The MCP-related nodes obtain the tool metadata of AI tools based on the MCP protocol, construct a relationship graph between the tool shelf group and AI tools, and configure resource robot units for the tool shelf group. The cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into the listing candidate list and the delisting candidate list based on the shelf visibility score; The cloud disk server node responds to the user terminal's AI tool access request by displaying the list of candidate listings to the user terminal. The cloud disk server node responds to the AI ​​tool selection request from the user terminal, assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution, and triggers the listing or delisting event of the AI ​​tool based on the real-time monitored resource usage.

16. The management system for the AI ​​tool as described in claim 15, characterized in that, The MCP-related nodes include MCP client nodes, MCP server nodes, and MCP coordinator nodes; the system also includes tool metadata database nodes and resource robot nodes. The MCP-related nodes then obtain the tool metadata of the AI ​​tools based on the MCP protocol, construct a relationship graph between the tool shelf group and the AI ​​tools, and configure resource robot units for the tool shelf group, specifically: The MCP client node submits the tool metadata of the AI ​​tool to be registered; The MCP server node parses the tool metadata, generates a tool context object, and stores the tool context object in the tool metadata database node; The MCP coordinator node reads the tool context object from the tool meta-database node, executes a multi-loop fuzzy attribution algorithm to dynamically optimize the structure of the tool shelf group, assigns a corresponding tool shelf group to the AI ​​tool, generates a relationship graph between the tool shelf group and the AI ​​tool, and stores it in the tool meta-database node; The MCP coordinator node binds resource robot units to the updated tool shelf group according to the relationship graph, and sends resource allocation initialization instructions to the resource robot node; The resource robot node configures resource robot units for the tool shelf group according to the resource allocation initialization command.

17. The management system for the AI ​​tool as described in claim 16, characterized in that, The tool metadata includes tool function vectors, tool runtime resource requirements, and tool runtime mode identifiers; Wherein, the tool function vector represents the tool operation characteristics of the AI ​​tool, the tool operation resource requirement represents the operation resource requirement of the AI ​​tool, and the tool operation mode identifier includes the priority category information of the AI ​​tool; The tool context object includes a unique tool identifier, a tool function vector, tool runtime resource requirements, and a tool runtime mode.

18. The management system for the AI ​​tool as described in claim 16, characterized in that, The MCP coordinator node is specifically used for: Initialize the soft attribution coefficient matrix; wherein, the elements of the soft attribution coefficient matrix represent the fuzzy coefficients by which the AI ​​tool is assigned to the tool shelf group; Initialize the center vector of each tool shelf group; wherein each tool shelf group represents an AI application type; Iterate through each AI tool and each tool shelf group, and calculate the Euclidean distance between the current AI tool and the current tool shelf group; Calculate the soft attribution coefficient based on the Euclidean distance; Based on the soft affiliation coefficient, the structure of the current tool shelf group is divided, merged, or maintained, and the corresponding tool shelf group is assigned to the current AI tool; Based on the current attribution results of the AI ​​tools, dynamically update the center vector of all tool shelf groups; When the change in the center vector of all tool shelf groups is less than a preset change threshold in a first preset number of iterations, or when the number of iterations reaches a second preset number, the iteration process is terminated and the structure of all tool shelf groups is locked; wherein, the second preset number is greater than the first preset number. The AI ​​tools and the tool shelf groups are mapped to generate a relationship graph between the tool shelf groups and the AI ​​tools, and stored in the tool metadata database node; The step of dividing, merging, or maintaining the structure of the current tool shelf group based on the soft affiliation coefficient, and assigning a corresponding tool shelf group to the current AI tool, includes: When the soft affiliation coefficient is greater than a preset coefficient threshold, the current AI tool is affixed to the current tool shelf group; When the soft affiliation coefficient is less than or equal to a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is less than or equal to a preset number threshold, the current tool shelf group is divided into a first sub-tool shelf group and a second sub-tool shelf group, and the current AI tool is affixed to the first sub-tool shelf group or the second sub-tool shelf group. When the soft attribution coefficient is greater than a preset coefficient threshold, and the number of tool shelves in the current tool shelf group is greater than a preset number threshold, the current tool shelf group is merged with other tool shelf groups with the highest similarity to obtain a combined tool shelf group, and the current AI tool is assigned to the combined tool shelf group.

19. The management system for the AI ​​tool as described in claim 16, characterized in that, The MCP coordinator node is also used for: Obtain the relationship graph between the tool shelf group and AI tools from the tool meta-database node, determine the set of AI tools corresponding to the tool shelf group, and extract the tool operation resource requirements of each AI tool; Calculate the resource requirement vector of the tool shelf group based on the tool operation resource requirements of the AI ​​tools in the tool shelf group; Calculate the computational flow density of the tool shelf group based on the resource demand vector; Based on the computation flow density and resource demand vector of the tool shelf group, select and bind a resource robot unit from the resource robot unit pool; Based on the binding relationship between the tool shelf group and the resource robot unit, a resource allocation initialization command is generated and sent to the resource robot node; The resource robot units include lightweight, standard, heavy-duty, and dynamic types; the dynamic resource robot nodes are connected to the elastic resource pool and can dynamically apply for and release resources. The step of selecting and binding a resource robot unit from the resource robot unit pool based on the computation flow density and the resource demand vector of the tool shelf group includes: When the computation flow density is less than or equal to a preset first computation flow density threshold, a lightweight resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset first computation flow density threshold and less than or equal to a preset second computation flow density threshold, a standard-level resource robot unit is selected from the resource robot unit pool for binding. When the computation flow density is greater than a preset second computation flow density threshold, a heavyweight resource robot unit is selected from the resource robot unit pool for binding; the second computation flow density threshold is greater than the first computation flow density threshold. When the variance of the resource demand vector is greater than a preset variance threshold, a dynamic resource robot unit is selected from the resource robot unit pool for binding.

20. The management system for the AI ​​tool as described in claim 16, characterized in that, The resource robot node is specifically used for: In response to the resource allocation initialization command, a data structure for the corresponding resource robot unit is created for the tool shelf group; Read parameter data of all AI tools in the current tool shelf group from the tool metadata database node, and update its own task list and scheduling queue; The binding status and resource quota information are fed back to the tool's metadata database node in real time.

21. The management system for the AI ​​tool as described in claim 15, characterized in that, The cloud disk server node evaluates the shelf visibility score of the AI ​​tool in real time, and classifies the AI ​​tool into the listing candidate list and the delisting candidate list based on the shelf visibility score, specifically as follows: The cloud disk server node calculates the AI ​​supermarket demand score for each AI tool based on the relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score. The cloud disk server node sorts the shelf visibility scores in descending order, adds the top K AI tools with shelf visibility scores greater than or equal to a preset first visibility score threshold to the list of candidates for listing, and adds AI tools with shelf visibility scores less than a preset second visibility score threshold to the list of candidates for delisting; wherein, the first visibility score threshold is greater than the second visibility score threshold.

22. The management system for the AI ​​tool as described in claim 21, characterized in that, The cloud disk server node calculates an AI supermarket demand score for each AI tool based on relevant parameters of the AI ​​tool, using a time window and consensus binding algorithm, including: By counting the total number of times the AI ​​tool was invoked by users within the current time window; The corresponding weighting coefficients are determined based on the type of AI tool; wherein, the weighting coefficients are used to distinguish the degree of influence of different types of AI tools on the AI ​​supermarket demand score; The degree of association between the AI ​​tool and the tool shelf group is calculated, and a weighted coefficient is obtained by weighting using a normalized soft attribution coefficient. The AI ​​supermarket demand score is obtained by multiplying the total number of times the AI ​​tool is called by the user, the weight coefficient, and the weighting coefficient. The cloud disk server node generates a shelf visibility score for each AI tool based on the AI ​​supermarket demand score, including: Calculate the load status parameters based on the computational flow density of the tool shelf group to which the AI ​​tool belongs; Calculate the normalized Euclidean distance based on the tool function vector of the AI ​​tool and the center vector of the tool shelf group to which it belongs; The shelf visibility score of the AI ​​tool is calculated based on the AI ​​supermarket demand score, the load condition parameter, and the normalized Euclidean distance.

23. The management system for the AI ​​tool as described in claim 16, characterized in that, The cloud disk server node responds to the user terminal's AI tool selection request by assigning the target AI tool in the request to the corresponding resource robot unit for execution, and triggers an add-on or remove-off event for the AI ​​tool based on real-time monitored resource usage, including: In response to the AI ​​tool selection request from the user terminal, the cloud disk server node assigns the target AI tool in the AI ​​tool selection request to the corresponding resource robot unit for execution; The resource robot node monitors the resource utilization rate of the target AI tool in real time; When the resource utilization rate of the target AI tool exceeds a preset first resource utilization rate threshold, the resource robot node sends a delisting signal to the cloud disk server node. In response to the delisting signal, the cloud disk server node updates the shelf visibility score of the target AI tool, marks the target AI tool as temporarily delisted, and removes it from the list of listing candidates. When the resource utilization rate of the target AI tool is less than a preset second resource utilization rate threshold, the resource robot node sends an uploading signal to the cloud disk server node; the second resource utilization rate threshold is less than the first resource utilization rate threshold. In response to the listing signal, the cloud disk server node updates the shelf visibility score of the target AI tool and reclassifies the target AI tool into the listing candidate list; The cloud disk server node counts the cumulative number of times the target AI tool has been removed from the platform. When the cumulative number of removals exceeds a third preset number, the target AI tool is added to the removal candidate list.

24. A management device for an AI tool, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a management method for the AI ​​tool as described in any one of claims 1 to 14.

25. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform a management method for the AI ​​tool as described in any one of claims 1 to 14.

26. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, implement the management method of the AI ​​tool as described in any one of claims 1 to 14.

Images