Multi-agent cooperation method and device based on MCP and medium

Abstract

The invention discloses an MCP-based multi-agent collaboration method and device and a medium, and relates to the technical field of artificial intelligence, the method comprises the following steps: obtaining a model registry form through a standardized model registration interface, analyzing model metadata in the registry form, and extracting a core field of a current registration model; creating a model node in the model map, and initializing the model node based on the core field; receiving a task request, triggering a scheduling engine, determining a plurality of candidate model nodes in the model atlas according to task content, respectively calculating task execution scores of the candidate model nodes, and generating a model execution chain based on the task execution scores; and based on the model execution chain, scheduling a candidate model corresponding to the candidate model node, and executing the corresponding task content. Dynamic access of multiple types of models is supported through a standardized model registration interface, system expansibility and integration efficiency are improved, and matching precision of tasks and models and multi-model cooperation continuity are enhanced.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a multi-agent collaboration method, device, and medium based on MCP. Background Art

[0002] Multi-Model Collaboration (MCP) is a revolutionary software development paradigm designed to efficiently complete complex development tasks through the collaborative work of multiple agents. Its core concept is to decompose complex development tasks into specialized agents, each responsible for specific responsibilities, and achieve efficient collaboration through standardized collaboration mechanisms. With the development of artificial intelligence technology, multi-model collaboration has become a key means of improving complex task processing capabilities.

[0003] There are some problems with existing model integration methods. On the one hand, model integration relies on customized interfaces. When adding or replacing models, other models in the entire model integration system need to be modified accordingly, which limits the scalability of the model. On the other hand, context information is stored in each model in a dispersed manner, making it difficult for different models to efficiently share and synchronize context information when processing tasks. In addition, model scheduling strategies are mostly based on static rules, which are difficult to adapt to dynamic task requirements and model state changes, seriously restricting the scalability and intelligence level of the intelligent agent system. Summary of the Invention

[0004] To solve the above problems, this application proposes a multi-agent collaboration method based on MCP, including:

[0005] Obtain the model registration form through the standardized model registration interface, parse the model metadata in the registration form, and extract the core fields of the current registered model;

[0006] Creating a model node of the current registered model in the model graph, and initializing node attributes and node associations of the model node based on the core fields to complete the registration process of the current registered model;

[0007] Receive a task request, trigger the scheduling engine, determine multiple candidate model nodes in the model graph according to the task content, calculate the task execution scores of the candidate model nodes respectively, and generate a model execution chain based on the task execution scores;

[0008] Based on the model execution chain, the candidate model corresponding to the candidate model node is scheduled to execute the corresponding task content.

[0009] On the other hand, the present application also proposes a multi-agent collaborative device based on MCP, comprising:

[0010] at least one processor; and,

[0011] a memory communicatively connected to the at least one processor; wherein,

[0012] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a multi-agent collaboration method based on MCP as described in the above example.

[0013] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as: a multi-agent collaboration method based on MCP as described in the above example.

[0014] This application proposes a multi-agent collaboration method based on MCP, which can bring the following beneficial effects:

[0015] Through the standardized model registration interface, it supports dynamic access to multiple types of models, improving system scalability and integration efficiency. It also realizes semantic management and association of model capabilities through graphical modeling, enhancing the matching accuracy between tasks and models and the collaborative coherence of multiple models.

[0016] Through multi-factor scheduling strategies, it comprehensively evaluates model capabilities, context, and load status, optimizes resource allocation, and improves task processing efficiency. Its abstract architecture can flexibly adapt to various AI scenarios such as dialogue systems and intelligent question-and-answer systems, and promote the evolution of artificial intelligence from single-model operation to deep collaborative multi-model operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 Schematic diagram of a multi-agent collaboration method based on MCP in an embodiment of the present application;

[0019] Figure 2 This is a schematic diagram of data compatible association edges in the model graph in the embodiment of the present application;

[0020] Figure 3 This is a schematic diagram of an MCP-based multi-agent collaborative device in an embodiment of the present application. DETAILED DESCRIPTION

[0021] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0022] The technical solutions provided by the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] like Figure 1 As shown, the embodiment of the present application provides a multi-agent collaboration method based on MCP, including:

[0024] S101: Obtain a model registration form through a standardized model registration interface, parse the model metadata in the registration form, and extract core fields of the current registered model.

[0025] Specifically, a model registration request is received, and a model registration form is obtained through a standardized model registration interface. The data format of the model registration form is converted into a standardized data format corresponding to a preset standard transmission protocol (such as JSON Schema or a custom template), and stored in the metadata database.

[0026] Before receiving model registration requests, a communication protocol layer is defined for a standardized model registration interface based on a pre-set standard transmission protocol. Examples include a RESTful interface or a gRPC interface. The RESTful interface is based on the HTTP protocol, uses methods such as POST or register to receive registration requests, and supports data formats such as JSON and XML. The gRPC interface is defined based on the HTTP / 2 protocol and the Protobuf protocol, enabling efficient binary transmission. Furthermore, the data standardization layer of the model registration interface is standardized through the MCP protocol specification, defining the metadata format for registration requests to ensure that all models submit uniform structured information.

[0027] Furthermore, the protocol parsing and verification layer of the standardized model registration interface parses the model metadata in the registration form, extracts the core fields of the currently registered model, and verifies the data legitimacy, including checking field formats, verifying the integrity of required fields, and restricting illegal values. Core fields include model version, input and output data, capability tags, call constraints, security authentication, and context.

[0028] It should be noted that the model version field includes the model name and version information; the input and output data field includes the input and output data formats (such as text, images, structured data); the capability label field includes the model capability label (such as sentiment analysis, entity recognition, image classification, etc.); the call constraint field includes the call constraints (such as the maximum number of concurrency, processing delay, resource consumption level); the security authentication field includes security authentication information (such as API keys, signature mechanisms); the context field includes supported context types (task level, session level, user level, etc.).

[0029] S102: creating a model node of the current registered model in the model graph, and initializing the node attributes and node associations of the model node based on the core fields to complete the registration process of the current registered model.

[0030] Specifically, a model node of the currently registered model is created in the model graph, the core fields are added to the node attributes of the created model node, and a multi-dimensional association edge is constructed between the model node and the registered nodes in the model graph based on the core fields. Among them, the multi-dimensional association edge includes data compatibility edge, capability similarity edge, version inheritance edge, context adaptation edge, etc. Figure 2 shown.

[0031] Among them, for data compatible edges, based on the input and output data fields, the current model input and output formats of the current registered model are determined, all registered model nodes in the graph database are traversed, and the model input and output formats corresponding to the registered model nodes are matched with the current model input and output formats for compatibility. According to the compatibility matching results, data compatible edges corresponding to the model nodes and the registered model nodes are constructed respectively.

[0032] It should be noted that the compatibility matching process includes input compatibility and output compatibility. Input compatibility checks whether the output schema of the current model matches the input schema of the model corresponding to the registered model node (including data format and field types). Output compatibility checks whether the input schema of the current model matches the output schema of the model corresponding to the registered model node. For example, if the output schema of model A is compatible with the input schema of model B, a data compatibility edge is established between models A and B. The edge attributes can record the compatible schema types (for example, {"format":"json","fields":["text"]}).

[0033] For capability similarity edges, obtain the existing model capability vector corresponding to the registered model node, calculate the second similarity between the current model capability vector and the existing model capability vector, and construct capability similarity edges corresponding to the model node and the registered model node based on the second similarity. In this embodiment of the present application, if the second similarity exceeds the threshold, a capability similarity edge is established, and the edge attribute records the similarity score (e.g., {"score":0.85}).

[0034] For the version inheritance edge, according to the current model version information, the historical version model node corresponding to the current registered model in the model graph is determined, and according to the compatibility matching result corresponding to the historical version model node and the model node, it is determined whether to construct a version inheritance edge between the historical version model node and the model node. In the embodiment of the present application, the version identification matching is first performed to check whether the current registered model name or metadata contains the identifier of the old version. If it does, a compatibility verification is performed to ensure that the input / output Schema of the new version is compatible with the old version. If there are major changes (such as format incompatibility, etc.), it is marked as not inherited; if it is a compatible upgrade after the compatibility verification, a version inheritance edge is established between the historical version model node and the model node, and the edge attribute records the version number difference.

[0035] For context adaptation edges, obtain the pre-built context template in the model graph, match the current model context support type pre-context template, determine the context adaptation template, determine the registered model node corresponding to the context adaptation template, and build a context adaptation edge with the model node. In an embodiment of the present application, a standard context template is pre-defined, the context support type declared when the current registered model is registered is extracted, and a matching context template node is searched in the model graph to establish a context adaptation edge, indicating that the model is adapted to the context structure.

[0036] Furthermore, after completing the registration process of the currently registered model, the model graph is updated, and the model graph index cache is refreshed (such as writing to Redis / Memcached) to facilitate subsequent scheduling to quickly find matching models.

[0037] In an embodiment of the present application, before initializing the node attributes and node associations of the model nodes based on the core fields, the model capabilities of the current registered model are determined based on the capability label field, the standard capability vocabulary is obtained, the first similarity between each vocabulary node in the standard capability vocabulary and the model capability is calculated, multiple similar vocabulary nodes are determined, the similar vocabulary nodes are vectorized, and weighted based on the first similarity to obtain the current model capability vector of the current registered model.

[0038] It's important to note that standard vocabulary nodes are predefined, structured capability classifications, while actual capability labels may be more diverse and non-uniform. For example, different models may use different terms to describe similar capabilities. For example, "emotion recognition" and "sentiment analysis" may refer to the same capability. Directly vectorizing these different labels may result in discrepancies between the vectors, even if their capabilities are actually similar. By mapping different capability labels to a unified standard vocabulary node—for example, mapping "emotion recognition" and "sentiment analysis" to the standard node "sentiment analysis"—this terminology difference can be eliminated, making the capability descriptions of different models more consistent. The standard vocabulary nodes are then vectorized and the capability label vectors are synthesized based on the mapping weights. This maintains semantic consistency while preserving the subtle differences in the specific capabilities of each model.

[0039] S103: Receive a task request, trigger the scheduling engine, determine multiple candidate model nodes in the model graph according to the task content, calculate the task execution scores of the candidate model nodes respectively, and generate a model execution chain based on the task execution scores.

[0040] Specifically, the user task request is received through the API gateway, and the scheduling engine is triggered to obtain the task content, parse the task content, extract the key information of the task, execute the multi-factor scheduling algorithm based on the key information of the task, and select the model based on the comprehensive multi-factor factors.

[0041] Key task information includes task category, raw input data, task context, and performance constraints. Based on this information, a structured query statement is generated and queried within the model graph to identify multiple candidate model nodes.

[0042] In an embodiment of the present application, the execution process of the multi-factor scheduling algorithm includes: calculating the task execution scores of multiple candidate model nodes, including context scores, model capability matching scores and load scores.

[0043] Based on the context score, determine the candidate context template corresponding to the candidate model node, use the pre-trained semantic model to vectorize the task context and the candidate context template, calculate the third similarity corresponding to the task context and the candidate context template, and obtain the context score of the candidate model node based on the third similarity. If the third similarity is lower than the threshold, directly exclude the model.

[0044] For the model capability matching score, obtain the candidate model capability vector corresponding to the candidate model node, vectorize the task type, calculate the fourth similarity between the task type and the candidate model capability vector, and obtain the model capability matching score of the candidate model node based on the fourth similarity.

[0045] For the load score, the heartbeat mechanism is used to obtain the model load status data corresponding to the candidate model node, and the load score corresponding to the candidate model node is calculated based on the model load status data.

[0046] In addition, in this embodiment, risky models are filtered through security and permission policies to ensure data security. Specifically, the task initiator is checked to see if they have the authority to call the target model, and all models that do not comply with the ACL are filtered before scheduling, ensuring that only legitimate models enter the score calculation phase.

[0047] Furthermore, it is determined whether the task request corresponds to a single-model task or a multi-model task. When it is determined to be a single-model task, the context score, model capability matching score and load score are weighted based on the preset weight coefficient to obtain the comprehensive score corresponding to the candidate model node.

[0048] Determine whether the task request corresponds to a single-model task or a multi-model task. When it is determined to be a single-model task, directly call the model with the highest comprehensive score to execute the single-model task; when it is determined to be a multi-model task, sort the candidate model nodes according to the comprehensive score, and determine the feasible links based on the data compatible edges of the candidate model nodes in the model graph. Based on the order of the candidate model nodes and the feasible links, build a model execution chain.

[0049] S104: Based on the model execution chain, scheduling the candidate model corresponding to the candidate model node to execute the corresponding task content.

[0050] Specifically, the task request is encapsulated as an MCP protocol standard message, and based on the model execution chain, it is sent to the candidate model, the corresponding task content is executed, the execution result output by the candidate model is obtained, the execution result is aggregated, and returned to the initiator of the task request.

[0051] In an embodiment of the present application, the original task request is encapsulated into a standard message according to the MCP protocol, including input data, context identifier and link configuration. The task is distributed to the first model node in link order through the service bus. After each model node is processed, the output result is converted into the input format of the next node according to the MCP protocol, the intermediate result is written to the shared context, and consistency is guaranteed by version number or transaction lock. If a node fails to process, the rollback mechanism is triggered, the dirty data is cleared and the upstream node is notified to terminate execution.

[0052] Furthermore, the final output result will be written to the final_output field in the context structure and returned to the task initiator through the API. Supported return methods include: synchronous call, where the return result includes a status code and processed data; asynchronous notification, where the platform calls a callback address (webhook) for notification; and multi-round dialogue, where the return result is retained in the context for the next round of model processing.

[0053] This application supports dynamic access to multiple types of models through a standardized model registration interface, improving system scalability and integration efficiency. It also implements semantic management and association of model capabilities through graphical modeling, enhancing the matching accuracy between tasks and models and the collaborative coherence of multiple models.

[0054] Through multi-factor scheduling strategies, it comprehensively evaluates model capabilities, context, and load status, optimizes resource allocation, and improves task processing efficiency. Its abstract architecture can flexibly adapt to various AI scenarios such as dialogue systems and intelligent question-and-answer systems, and promote the evolution of artificial intelligence from single-model operation to deep collaborative multi-model operation.

[0055] like Figure 2 As shown, the embodiment of the present application also proposes a multi-agent collaborative device based on MCP, including:

[0056] at least one processor; and,

[0057] a memory communicatively connected to the at least one processor; wherein,

[0058] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a multi-agent collaboration method based on MCP as described in any of the above embodiments.

[0059] An embodiment of the present application further provides a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to be: a multi-agent collaboration method based on MCP as described in any of the above embodiments.

[0060] The various embodiments in this application are described in a progressive manner. Similar portions between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the device and medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simple. For relevant portions, refer to the descriptions of the method embodiments.

[0061] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects to their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0062] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0064] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0066] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0067] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0068] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0069] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0070] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A multi-agent collaboration method based on MCP, characterized in that: include: Obtain the model registration form through the standardized model registration interface, parse the model metadata in the registration form, and extract the core fields of the current registered model; Creating a model node of the current registered model in the model graph, and initializing node attributes and node associations of the model node based on the core fields to complete the registration process of the current registered model; Receive a task request, trigger the scheduling engine, determine multiple candidate model nodes in the model graph according to the task content, calculate the task execution scores of the candidate model nodes respectively, and generate a model execution chain based on the task execution scores; Based on the model execution chain, the candidate model corresponding to the candidate model node is scheduled to execute the corresponding task content.

2. The multi-agent collaboration method based on MCP according to claim 1, characterized in that: The model registration form is obtained through the standardized model registration interface, specifically including: Define a standardized model registration interface based on the preset standard transmission protocol; A model registration request is received, and a model registration form is obtained through the standardized model registration interface to convert a data format of the model registration form into a standardized data format corresponding to the preset standard transmission protocol.

3. The multi-agent collaboration method based on MCP according to claim 1, characterized in that: The core fields include a model version field, an input and output data field, a capability tag field, a call constraint field, a security authentication field, and a context field; Before initializing the node attributes and node associations of the model nodes based on the core fields, the method further includes: Determining the model capability of the currently registered model based on the capability tag field; Obtaining a standard capability vocabulary, calculating a first similarity between each vocabulary node in the standard capability vocabulary and the model capability, and determining a plurality of similar vocabulary nodes; The similar vocabulary nodes are vectorized and weighted based on the first similarity to obtain a current model capability vector of the current registered model.

4. The multi-agent collaboration method based on MCP according to claim 3, characterized in that: Initializing the node attributes and node associations of the model nodes based on the core fields specifically includes: Based on the core fields, determine the current model input and output format, the current model capability vector, the current model version information, and the current model context support type of the current registered model, and add them to the node attributes of the model node; Traversing the registered model nodes in the model graph; Compatibility matching is performed on the model input and output formats corresponding to the registered model nodes and the current model input and output formats, and based on the compatibility matching results, data compatible edges are constructed between the model nodes and the registered model nodes; Obtaining an existing model capability vector corresponding to the registered model node, respectively calculating a second similarity between the current model capability vector and the existing model capability vector, and constructing capability similarity edges between the model node and the registered model node based on the second similarity; Determine, according to the current model version information, a historical version model node corresponding to the currently registered model in the model graph; Determining whether to construct a version inheritance edge between the historical version model node and the model node according to a compatibility matching result corresponding to the historical version model node and the model node; Obtaining a pre-built context template in the model graph, matching the current model context support type with the context template, and determining a context adaptation template; A registered model node corresponding to the context adaptation template is determined, and a context adaptation edge is constructed with the model node.

5. The multi-agent collaboration method based on MCP according to claim 1, characterized in that: Determining multiple candidate model nodes in the model graph according to the task content specifically includes: Obtaining task content, parsing the task content, and extracting key task information; the key task information includes task category, original input data, task context, and performance constraints; Based on the mission-critical information, a structured query statement is generated, and a query is performed in the model graph to determine a plurality of candidate model nodes.

6. The multi-agent collaboration method based on MCP according to claim 4 or claim 5, characterized in that: Calculating the task execution score of the candidate model node specifically includes: Determine a candidate context template corresponding to the candidate model node, and vectorize the task context and the candidate context template; Calculating a third similarity between the task context and the candidate context template, and obtaining a context score for the candidate model node based on the third similarity; Obtaining a candidate model capability vector corresponding to the candidate model node, vectorizing the task type, calculating a fourth similarity between the task type and the candidate model capability vector, and obtaining a model capability matching score for the candidate model node based on the fourth similarity; Obtain model load status data corresponding to the candidate model node, and calculate the load score corresponding to the candidate model node based on the model load status data.

7. The multi-agent collaboration method based on MCP according to claim 6, characterized in that: The generating model execution chain based on the task execution score specifically includes: Based on a preset weight coefficient, the context score, the model capability matching score, and the load score are weighted to obtain a comprehensive score corresponding to the candidate model node; Sorting the candidate model nodes according to the comprehensive scores; Based on the data compatible edges of the candidate model nodes in the model graph, a feasible link is determined, and based on the order of the candidate model nodes and the feasible link, a model execution chain is constructed.

8. The multi-agent collaboration method based on MCP according to claim 1, characterized in that: The scheduling of the candidate model corresponding to the candidate model node based on the model execution chain to execute the corresponding task content specifically includes: Encapsulate the task request into an MCP protocol standard message, and send it to the candidate model based on the model execution chain to execute the corresponding task content; The execution results output by the candidate models are obtained, the execution results are aggregated, and returned to the initiator of the task request.

9. A multi-agent collaborative device based on MCP, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a multi-agent collaboration method based on MCP as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are set to: a multi-agent collaboration method based on MCP as described in any one of claims 1 to 8.

Images