MCP integrated management system and method supporting multi-task cooperation

Abstract

The invention relates to the field of MCP protocol integrated management, in particular to an MCP integrated management system and method supporting multi-task collaboration, and the method comprises the steps: completing the deployment of one or more service instances according to the communication mode of an MCP Server through employing a corresponding deployment strategy, and registering the metadata of the MCP Server and the information of the service instances into a resource library; creating a personal cooperation space and a team cooperation space, and configuring role permissions for members of the team cooperation space; generating a workflow definition containing task execution logic; according to the workflow definition, creating and executing a corresponding workflow instance, and feeding back an execution state in real time; receiving a calling request for the managed MCP Server, and carrying out load balancing routing and protocol conversion to realize seamless calling of the third-party MCP service; the requirements in different scenes are met, and the flexibility and efficiency of cooperation are improved.

Description

Technical Field

[0001] This application relates to the field of MCP protocol integration and multi-task collaboration management, specifically to an MCP integration management system and method that supports multi-task collaboration. Background Technology

[0002] MCP (Model Context Protocol) serves as a unified communication protocol between models and external tools. Through standardized interaction specifications, it aims to resolve issues such as differences in syntax rules among large models and incompatibility of tool interfaces, enabling large models to directly call tools that conform to the MCP specification (such as weather queries and map services). MCP Server, as a standardized encapsulation service for external tools, supports two communication modes: STDIO and SSE, suitable for local inter-process communication and remote HTTP long connections, respectively.

[0003] While the MCP protocol provides general capabilities for tool invocation, significant shortcomings remain in practical integration and application. Currently, there is a lack of a unified management platform for MCP server registration, deployment, and version maintenance. Developers must manually adapt both STDIO and SSE modes, making the deployment process complex and requiring a high level of technical expertise, thus keeping the barrier to entry for MCP technology high. Furthermore, in multi-user collaborative task scenarios, existing technologies lack effective team collaboration mechanisms, failing to achieve task allocation, permission isolation, and execution progress tracking. Data conflicts easily arise when different users collaborate on editing, severely hindering overall team collaboration efficiency. Simultaneously, the invocation process of MCP tools often requires repetitive parameter configuration, resulting in low cross-platform inefficiency. It also lacks effective support for visual orchestration of complex task flows, failing to meet the needs of enterprise applications for process automation, monitorability, and reusability.

[0004] In summary, existing MCP technologies face a series of problems in practical implementation, such as complex integration and deployment, lack of unified lifecycle management, insufficient support for multi-user collaboration, and difficulty in task workflow orchestration. There is an urgent need for an integrated management solution that can achieve unified management of MCPServer, support real-time collaboration among multiple users, and provide visual workflow orchestration capabilities. Summary of the Invention

[0005] To address the above problems, this invention provides an MCP integrated management system and method that supports multi-task collaboration.

[0006] In a first aspect, the present invention provides an MCP integrated management system that supports multi-task collaboration, comprising: The MCP management module is used to receive MCP Server code packages or remote service addresses submitted by users, verify the protocol compatibility of the MCP Server, and after the verification is passed, deploy one or more service instances according to the communication mode of the MCP Server using the corresponding deployment strategy, and register the metadata of the MCP Server and the service instance information in the resource library to form an MCP Server list, so as to complete the management of the MCP Server. The multi-task collaboration space creation module is used to respond to user operations, create personal collaboration spaces and team collaboration spaces, and configure role permissions for members of the team collaboration space. The visual workflow orchestration module provides a visual drag-and-drop interface within the collaboration space. It receives drag-and-drop connection operations from users based on the MCP Servers in the MCP Server list to generate workflow definitions containing task execution logic. Based on the workflow definitions, it creates and executes corresponding workflow instances and provides real-time feedback on the execution status. The gateway service module is used to receive call requests to managed MCP servers, perform load balancing routing and protocol conversion, so as to achieve seamless calling of third-party MCP services; The visualization workflow orchestration module can access the resource library to obtain the MCP Server list, enabling workflow definitions generated based on the MCP Server list to directly call the managed MCP Servers through the gateway service module when executed.

[0007] As a further limitation of the technical solution of the present invention, the MCP management module includes: The code parsing unit is used to parse the code package submitted by the user in order to obtain the functional description, dependent environment and interface parameter metadata of the MCP Server. The protocol verification unit is used to verify the protocol compatibility of the MCP Server based on the MCP protocol standard. Specifically, for SSE communication mode, the protocol verification unit is configured to verify whether the MCP Server exposes the specified SSE path and returns data of type text / event-stream; for STDIO communication mode, the protocol verification unit is configured to verify whether the MCP Server supports standard input / output pipe communication. The deployment execution unit is used to adopt the corresponding deployment strategy according to the communication mode of the MCP Server; wherein, for the STDIO communication mode, the deployment execution unit is configured to start the service process in the local sandbox environment; for the SSE communication mode, the deployment execution unit is configured to build the code package into a container image and deploy it in the container orchestration platform. The registration index unit is used to store the metadata and service instance information of the successfully verified and deployed MCP Servers in the database, and to build an index in the cache according to function classification and communication mode to form a list of MCPServers in the resource library.

[0008] The protocol verification unit performs precise verification for both SSE and STDIO communication modes, ensuring the protocol compatibility of the MCP Server and avoiding call failures due to protocol incompatibility, thus improving system reliability. The deployment and execution unit employs different deployment strategies based on the MCP Server's communication mode, enabling rapid service deployment and elastic scaling, improving resource utilization and system scalability. The registration and indexing unit stores the MCP Server's metadata and service instance information in the database and creates an index in the cache, forming a list of MCP Servers in the resource repository. This facilitates quick querying and retrieval of MCP Servers, improving overall system performance.

[0009] As a further limitation of the technical solution of the present invention, the multi-task collaboration space creation module is configured as follows: Create personal collaboration spaces, which are only visible and operable by the space creator; and create team collaboration spaces, which support adding multiple collaboration members through invitation; when multiple users are detected editing the same task flow simultaneously in the same team collaboration space, an operation conversion algorithm is used to resolve editing conflicts and achieve real-time collaborative editing. Configuring role permissions for members of the team collaboration space includes: The administrator has full management rights over the team space, including managing members and their permissions; Editors have the authority to create and edit task workflows within the team space; Viewers only have permission to view the task flow and results within the team space.

[0010] A clear distinction is made between personal and team collaboration spaces. Personal spaces guarantee user privacy and independence, allowing users to freely create and manage personal tasks. Team spaces support multi-user collaboration, allowing members to be added via invitation, facilitating collaborative editing and management of task workflows and meeting collaboration needs in different scenarios. An operation conversion algorithm resolves editing conflicts, enabling true real-time collaborative editing. This feature significantly improves team collaboration efficiency and reduces conflicts and chaos caused by multiple editors. Members with different roles have corresponding permissions based on their responsibilities and needs, ensuring data security and confidentiality and effectively avoiding the risks of permission abuse and data leakage.

[0011] As a further limitation of the technical solution of the present invention, the multi-task collaboration space creation module further includes: The permission synchronization unit is used to update the changed permission information to the cache in real time when the role permissions of members in the team space change, and make the session permissions of all currently logged-in users take effect immediately. The status notification unit is used to push status change events within the team collaboration space to the front-end interface in real time via a WebSocket connection; the status change events include new task creation, task status updates, or member changes.

[0012] This ensures the timeliness and accuracy of permission changes, avoiding user restrictions or errors caused by delayed permission updates, and improving system responsiveness and user experience. This real-time notification mechanism allows users to stay informed about the latest developments within the team collaboration space, facilitating timely communication and collaboration among members.

[0013] As a further limitation of the technical solution of the present invention, the visual drag-and-drop interface is configured to support users in defining at least one of the following task execution logics through drag-and-drop and connection operations: Serial arrangement is used to define the sequential execution relationship of tasks; Parallel orchestration is used to define multiple tasks that can be executed simultaneously; Conditional orchestration is used to determine subsequent execution branches based on the execution results of preceding tasks.

[0014] The visual drag-and-drop interface supports user-defined serial orchestration, parallel orchestration, and conditional orchestration, meeting the diverse needs of task processes in different business scenarios. Users can flexibly design the execution order, concurrency relationships, and conditional branches of tasks according to actual business requirements, enabling the workflow to more accurately reflect business logic and improving its adaptability and flexibility.

[0015] As a further limitation of the technical solution of the present invention, the visual workflow orchestration module includes: A definition generation unit is used to transform the user's drag-and-drop connection operation into a structured workflow definition file, which defines the MCP Server's call order, parameter mapping relationship, and condition rules; The process engine unit is used to receive the workflow definition file, parse and convert it into a standardized BPMN model, and hand it over to a task scheduling engine to create and execute the workflow instance; The execution monitoring unit is used to record execution logs and performance metrics during the execution of the workflow instance; and to pause, terminate, or circuit breaker operations on the executing workflow instance in response to user instructions or abnormal situations; wherein the circuit breaker operation includes: when a call to a certain MCP Server fails consecutively, automatically switching the request to a backup instance of the MCP Server or other MCP Servers that can perform the same function.

[0016] This system standardizes and normalizes workflows from design to execution, improving readability, maintainability, and scalability, and facilitating subsequent management and optimization. The circuit breaker mechanism effectively prevents single points of failure, enhancing the system's fault tolerance and stability.

[0017] As a further limitation of the technical solution of the present invention, the gateway service module includes: An ecosystem adaptation unit is used to synchronize the metadata of the MCPServer from at least one third-party MCP ecosystem platform through a scheduled task, and uniformly register it in the resource library for the visualization workflow orchestration module to orchestrate. The load balancing routing unit is used to route a request to a certain MCP Server to one of the healthy service instances according to a preset load balancing strategy when a call request to a certain MCP Server is received. If the MCP Server has multiple service instances deployed, the load balancing strategy includes a least connections strategy or a round-robin strategy.

[0018] Intelligent load balancing strategies can allocate requests reasonably, avoid overloading of individual service instances, improve the overall performance and stability of the system, and ensure efficient processing of call requests.

[0019] As a further limitation of the technical solution of the present invention, the gateway service module further includes: The high availability guarantee unit is used to periodically send heartbeat detection requests to the service instances of the registered MCP Server. If no response is received from a service instance for a preset number of consecutive times, it is marked as unhealthy and the unhealthy service instance is temporarily excluded during routing until it recovers to health.

[0020] It can promptly detect abnormal states of service instances and automatically troubleshoot and recover, effectively improving the high availability of the system and reducing call failures caused by service instance failures.

[0021] As a further limitation of the technical solution of the present invention, the gateway service module is configured to perform a unified security verification on the request before the routing call request. The verification content includes: the validity of the API Key; and / or, whether the access frequency of the IP address initiating the request is within the allowed range.

[0022] By limiting the frequency of IP address access, the system's security capabilities are enhanced.

[0023] Secondly, the technical solution of the present invention also provides an MCP integration management method that supports multi-task collaboration, including: The system receives user-submitted MCP Server code packages or remote service addresses, verifies the protocol compatibility of the MCP Server, and, after successful verification, deploys one or more service instances using the corresponding deployment strategy based on the MCP Server's communication mode. It also registers the MCP Server's metadata and service instance information in the resource library to form an MCP Server list, thereby completing the management of the MCP Server. In response to user actions, create personal collaboration spaces and team collaboration spaces, and configure role permissions for members of the team collaboration spaces; Within the collaborative space, a visual drag-and-drop interface is provided to receive drag-and-drop connection operations performed by users based on MCP Servers in the MCP Server list, in order to generate a workflow definition containing task execution logic; based on the workflow definition, a corresponding workflow instance is created and executed, and the execution status is fed back in real time. It receives call requests to managed MCP Servers, performs load balancing routing and protocol conversion to enable seamless calls to third-party MCP services; The workflow definitions orchestrated in the collaboration space can directly call the managed MCP Server when executed.

[0024] It automates and improves the efficiency of multi-task collaboration, effectively solving problems such as complex MCP integration and deployment, lack of team collaboration mechanism, and low cross-platform call efficiency in existing technologies, thereby improving the overall efficiency and quality of multi-task collaboration.

[0025] As can be seen from the above technical solutions, this application has the following advantages: It achieves unified management of the entire process of MCP Server from receiving code packages or service addresses to verification, deployment, and registration. Whether it is a local code package or a remote service, it can efficiently complete the management and form a clear list of MCP Servers. This solves the problems of existing technologies, such as the lack of a unified platform for MCP Server registration, deployment, and version maintenance, and the need for manual adaptation to dual modes. It meets the diverse needs of different users in individual and team collaboration scenarios, provides a foundation for multi-user collaborative task management, and effectively improves the flexibility and efficiency of collaboration.

[0026] Users can generate workflow definitions through simple drag-and-drop connections without writing code. The system provides real-time feedback on the execution status of workflow instances, reducing the difficulty of workflow orchestration, improving orchestration efficiency, and meeting the needs of complex business scenarios. The gateway service module features load balancing routing and protocol conversion capabilities, enabling seamless invocation of third-party MCP services, expanding the platform's ecosystem compatibility, and enhancing the overall scalability and usability of the system. Attached Figure Description

[0027] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A block diagram of a system provided in an embodiment of the present invention.

[0029] Figure 2 A flowchart of a method provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this application and in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0032] like Figure 1 As shown in the figure, an embodiment of the present invention provides an MCP integrated management system that supports multi-task collaboration, comprising: The MCP management module is used to receive MCP Server code packages or remote service addresses submitted by users, verify the protocol compatibility of the MCP Server, and after the verification is passed, deploy one or more service instances according to the communication mode of the MCP Server using the corresponding deployment strategy, and register the metadata of the MCP Server and the service instance information in the resource library to form an MCP Server list, so as to complete the management of the MCP Server. The multi-task collaboration space creation module is used to respond to user operations, create personal collaboration spaces and team collaboration spaces, and configure role permissions for members of the team collaboration space. The visual workflow orchestration module provides a visual drag-and-drop interface within the collaboration space. It receives drag-and-drop connection operations from users based on the MCP Servers in the MCP Server list to generate workflow definitions containing task execution logic. Based on the workflow definitions, it creates and executes corresponding workflow instances and provides real-time feedback on the execution status. The gateway service module is used to receive call requests to managed MCP servers, perform load balancing routing and protocol conversion, so as to achieve seamless calling of third-party MCP services; The visualization workflow orchestration module can access the resource library to obtain the MCP Server list, enabling workflow definitions generated based on the MCP Server list to directly call the managed MCP Servers through the gateway service module when executed.

[0033] In some embodiments, the MCP management module includes: The code parsing unit is used to parse the code package submitted by the user in order to obtain the functional description, dependent environment and interface parameter metadata of the MCP Server. It's worth noting that the system also includes an interface module that receives user-uploaded code packages and decompresses them to a temporary working directory. The code parsing unit scans the decompressed directory structure, searching for and reading specific configuration files (such as package.json for Node.js, requirements.txt, or pyproject.toml for Python) to obtain information such as project dependencies and entry points. It locates and parses the MCP Server's entry point script files (such as index.js and server.py), extracting interface parameters through static code analysis or by reading specific structured comments. The system integrates and maps the scattered information obtained above into a structured metadata model, which contains key information such as functional descriptions, dependency lists, and interface parameters. The final standardized metadata output is then passed to the subsequent protocol verification unit for compliance checks.

[0034] The protocol verification unit receives metadata transmitted by the code parsing unit and determines whether the MCPServer is using SSE mode or STDIO mode based on the communication mode identifier in the metadata (such as the "communication_mode" field configured in the code package and the characteristics of the remote SSE service address).

[0035] SSE mode verification: Send an access request to the MCP Server to verify whether the preset / sse path is exposed. If the path cannot be accessed, the protocol is determined to be incompatible.

[0036] Receive data returned from the / sse path, verify that the data type is text / event-stream, and if the data type does not match, record the error message and determine that the verification failed.

[0037] Check whether the returned SSE event stream contains necessary fields such as "event" and "data" to ensure that the event stream format conforms to the MCP protocol standard. If any field is missing, the verification fails.

[0038] STDIO mode verification: A temporary process is created locally to attempt to establish a standard input / output pipe connection with the MCP Server. If the connection cannot be established, the protocol is deemed incompatible.

[0039] The test command is sent to the MCP Server through the pipeline, and the returned result is received. The result format is verified to conform to the STDIO data interaction standard specified by the MCP protocol. If the returned result is messy or cannot be parsed, the verification is deemed to have failed.

[0040] The openapi-spec-validator tool is used to perform a secondary compliance verification of the validation process and generate a validation report. If the validation passes, the metadata and validation report are transmitted to the deployment execution unit; if the validation fails, detailed error information is fed back to the front end for users to modify and adjust.

[0041] The deployment execution unit receives the pass result from the protocol verification unit and determines the corresponding deployment strategy based on the communication mode of the MCP Server (SSE mode corresponds to containerized deployment, and STDIO mode corresponds to local sandbox deployment). STDIO mode deployment: Call the Docker interface to create an independent local sandbox container and configure the basic runtime environment within the container. Unzip the MCPServer code package to the specified directory of the sandbox container, start the service process using uvx (Python) or npx (Node.js) tools, and automatically inject environment variables (such as API Key, database connection information, etc., obtained from the configuration center).

[0042] Monitor the startup status of the service process. If the process fails to listen to the port normally within 10 seconds after startup, it is determined that the startup has failed, the sandbox container is destroyed and an error message is returned; if the startup is successful, the sandbox container ID and the service process ID are recorded.

[0043] SSE mode deployment: Using the BuildKit tool, a Docker image is built based on the MCP Server code package and the dependency information from the metadata. The image tag includes the service name and version number (extracted from the metadata). The built Docker image is pushed to the platform's private image repository. If the push fails, an error is reported after three retries. A deployment request is sent to the Kubernetes cluster, specifying the image address, resource quota (e.g., 1 CPU core, 1GB memory, recommended configuration based on metadata), and exposed port (e.g., 8000). The Kubernetes scheduler deploys the image to an idle node in the cluster. After deployment, the API gateway interface is called to register the SSE service with the gateway. The registration path is / mcp / servers / {service name} / sse, ensuring that the gateway can route to the SSE service.

[0044] Based on Flowable7, standardized deployment nodes are defined, and the execution status of each node is recorded in real time. If a node fails to execute, a rollback mechanism is triggered, and the failed node and the reason are fed back to the front end.

[0045] The registration index unit receives deployment success information (such as sandbox container ID, Kubernetes deployment instance ID, and service access address) from the deployment execution unit. It integrates the MCP Server's metadata (functional description, dependent environment, interface parameters) and service instance information (instance ID, access address, communication mode, deployment node) into structured data, and stores it in the MCP Server information table by operating the MySQL database through MyBatis Plus. It connects to the Redis database and builds a secondary index using the MCP Server's functional category (such as "Weather Service" or "Map Service") and communication mode (SSE / STDIO) as index keys. For example, the value for "Functional Category: Weather Service" is a list of service IDs for all MCP Servers under this category, and the value for "Communication Mode: SSE" is a list of service IDs for all SSE mode MCP Servers, enabling fast querying. Newly managed MCP Server information is added to the system resource library, updating the MCP Server list to ensure the front-end can obtain the latest service list in real time. Simultaneously, audit information such as registration time and registered user ID is recorded and stored in the registration log table. If data storage and index creation are both successful, a "MCP Server managed successfully" message is returned to the front end, along with basic information about the service instance (such as access address and communication mode). If any step fails (e.g., an abnormal Redis connection causes index creation to fail), a retry mechanism is triggered. If the retry fails after two attempts, an error message is returned, and the storage operation in the database is rolled back.

[0046] In some embodiments, the multi-task collaboration space creation module is configured as follows: Create personal collaboration spaces, which are only visible and operable by the space creator; and create team collaboration spaces, which support adding multiple collaboration members through invitation; when multiple users are detected editing the same task flow simultaneously in the same team collaboration space, an operation conversion algorithm is used to resolve editing conflicts and achieve real-time collaborative editing. Configuring role permissions for members of the team collaboration space includes: The administrator has full management rights over the team space, including managing members and their permissions; Editors have the authority to create and edit task workflows within the team space; Viewers only have permission to view the task flow and results within the team space.

[0047] After a user logs into the platform, the system automatically checks whether the user has created a personal collaboration space. If not, the personal space initialization process is triggered. When a user initiates the creation of a team space, the team space creation process is triggered.

[0048] Creating a personal collaborative space: Automatically generate a unique ID for the personal space (e.g., "user_123_space_456"), associate it with the user ID, set the space type to "private," and set the default permissions to "visible and operable only by the creator." Store the personal space metadata (space ID, user ID, space type, creation time) in the collaborative space information table via MyBatis Plus, and cache the space's permission configuration in Redis (e.g., "Space ID: user_123_space_456 → Permission: Private"). Push a message to the front end indicating successful personal space creation, guiding the user into their personal space.

[0049] Creating a team collaboration space: The front-end displays a team space creation form, receiving user input for the space name, space description, and initial member invitation list. It automatically generates a unique team space ID, associates it with the creator's user ID, sets the space type to "team," and defaults to an administrator role for the creator. The system validates the entered corporate email address; upon successful validation, it generates an invitation link containing the space ID and invitation code, valid for 24 hours. Invitation emails are then sent to each member's email address via the platform's email service, containing the invitation link, space name, and creator information.

[0050] Store team space metadata (space ID, creator ID, space name, space description, creation time) in the collaborative space information table, and store initial member invitation records (email, invitation link, invitation status "pending confirmation") in the space member invitation table. Assign the "administrator" role (full permissions) to the creator, record the mapping relationship of "space ID-user ID-role-permission list" in the permission configuration table, and synchronize it to the Redis cache.

[0051] A real-time connection between the front-end and back-end is established via WebSocket. Once the space is created, a message indicating successful space creation and basic space information are pushed to the front-end in real time to ensure timely updates to the front-end interface.

[0052] When a team space is created, the default creator is the administrator. When a member joins the space via an invitation link, the default role is viewer. The administrator can modify member roles in the member management interface. When an administrator initiates a role change operation on the front end, the front end sends the "space ID - target user ID - new role" information to the back end. After receiving the role change request, the back end updates the corresponding role-permission mapping relationship in the permission configuration table and updates the permission information in the Redis cache using the SaToken tool to ensure that the permission change takes effect in real time. If the change fails (e.g., the target user does not exist), an error message is returned to the front end. When a user performs an operation (such as editing a task or deleting a workflow) within the team space, the front end first retrieves the user's role permissions in the current space from the Redis cache to determine if the user has the necessary permissions. If no permission information is found in the cache, a permission query request is sent to the back end. The back end queries the permission configuration table and returns the result. The front end then allows or denies the user's operation based on the result.

[0053] In this embodiment of the invention, the multi-task collaboration space creation module further includes: The permission synchronization unit is used to update the changed permission information to the cache in real time when the role permissions of members in the team space change, and make the session permissions of all currently logged-in users take effect immediately. The backend listens for changes to the permission configuration table (such as table data updates caused by role assignment or role change operations). When a change is detected, the permission synchronization process is triggered.

[0054] Remove the old permission information corresponding to the space ID-user ID from the Redis cache to avoid stale cached data. Based on the updated role-permission mapping in the permission configuration table, write the new permission information to the Redis cache, setting the cache expiration time to 24 hours and automatically refreshing upon expiration. If the user is currently logged in and has entered the corresponding team space, update the user's current session permission information using the SaToken tool to ensure the user can use the new permissions without logging in again. The backend periodically compares the permission configuration table with the permission information in the Redis cache; if inconsistencies are found, trigger a forced synchronization process to ensure data consistency.

[0055] The status notification unit is used to push status change events within the team collaboration space to the front-end interface in real time via a WebSocket connection; the status change events include new task creation, task status updates, or member changes.

[0056] In some embodiments, the visual drag-and-drop interface is configured to support users in defining at least one of the following task execution logics through drag-and-drop and connection operations: Serial arrangement is used to define the sequential execution relationship of tasks; Parallel orchestration is used to define multiple tasks that can be executed simultaneously; Conditional orchestration is used to determine subsequent execution branches based on the execution results of preceding tasks.

[0057] The visualization workflow orchestration module includes: Define the generation unit, which is used to transform the user's drag-and-drop connection operation into a structured workflow definition file. This file defines the MCP Server's call order, parameter mapping relationship and condition rules. The front-end is built with the Vue-Dragula component to create a visual drag-and-drop interface. The left side of the interface displays a list of MCPServers in the resource library (obtained from Redis cache or MySQL database), with each MCPServer presented as an icon and name. The middle of the interface is a canvas area for users to orchestrate workflows. The right side of the interface is a property configuration panel for configuring the parameters, conditions, and rules of the selected MCPServer.

[0058] Users drag and drop the target MCP Server icon from the MCP Server list on the left to the canvas area. The front-end records the service ID and location coordinates of the MCP Server and generates the corresponding task node on the canvas. Users click on the output of one task node and drag it to the input of another task node. The front-end records the connection relationship between the two task nodes and determines the orchestration type (serial, parallel, conditional). For example, a single connection between two nodes is a serial orchestration; multiple nodes connected to the same subsequent node simultaneously is a parallel orchestration; configuring conditional branching parameters when connecting nodes is a conditional orchestration. In the property configuration panel, users configure interface parameters and execution timeout for each task node; for conditional orchestration nodes, they configure conditional judgment rules, and the front-end associates the configuration information with the task nodes.

[0059] The front-end generates a JSON-formatted workflow definition file based on the user's drag-and-drop connection operations, parameter configurations, and rule configurations. The file contains information such as workflow ID, name, list of task nodes (each node includes service ID, parameters, and timeout), node connection relationships (orchestration type and conditional rules), creator ID, and space ID.

[0060] The frontend performs syntax validation on the JSON-formatted workflow definition to ensure there are no formatting errors. If the validation passes, the workflow definition is sent to the backend, which stores it in the workflow definition table and caches the workflow definition (associated with the space ID) in Redis for subsequent execution. If the validation fails, an error message (such as "conditional rule missing 'otherwise' branch") is sent back to the frontend.

[0061] The process engine unit is used to receive the workflow definition file, parse and convert it into a standardized BPMN model, and hand it over to a task scheduling engine to create and execute the workflow instance; Receive workflow execution requests (including workflow ID and space ID) sent by the front end, retrieve the corresponding JSON format workflow definition from the Redis cache or workflow definition table, and parse the JSON definition into a standardized BPMN model through the API provided by Flowable7, ensuring that the model conforms to the BPMN 2.0 specification.

[0062] Based on the parsed BPMN model, the Flowable7 process instance creation interface is called to generate a unique workflow instance ID, associate the workflow ID, space ID, and executing user ID, record the instance creation time and initial state ("to be executed"), and store them in the workflow instance table.

[0063] The workflow instance information is sent to the XXL-JOB task scheduling engine, specifying the scheduling strategy (such as immediate execution or scheduled execution). XXL-JOB generates a task execution plan based on the task node order and orchestration type in the workflow instance, and schedules the MCP Server call requests corresponding to each task node sequentially or in parallel.

[0064] Receive real-time feedback on task execution status from XXL-JOB (such as "Executing", "Execution Successful", "Execution Failed") and update the status of the corresponding task node in the workflow instance table.

[0065] When a task node fails to call the MCP Server consecutively (e.g., all three calls time out), the Sentinel circuit breaker mechanism is triggered, suspending the execution of the task node. The execution is then switched to a backup MCP Server instance for re-execution according to a preset policy, or the user is notified of the execution failure and asked to intervene.

[0066] After all task nodes of the workflow instance have been executed (regardless of success or failure), the execution results (such as the execution status of each node, output data, and reasons for failure) are compiled into a structured report, stored in the workflow execution result table, and pushed to the front end in real time via WebSocket for users to view.

[0067] The execution monitoring unit is used to record execution logs and performance metrics during the execution of the workflow instance; and to pause, terminate, or circuit breaker operations on the executing workflow instance in response to user instructions or abnormal situations; wherein the circuit breaker operation includes: when a call to a certain MCP Server fails consecutively, automatically switching the request to a backup instance of the MCP Server or other MCP Servers that can perform the same function.

[0068] The real-time feedback on execution status includes: displaying the overall execution progress of the workflow instance and the current status of each task in a graphical manner on the front-end interface in real time.

[0069] The backend parses JSON using Flowable7, transforms it into a standardized BPMN model, and schedules its execution via XXL-JOB. During execution, Logback records key information (call time, parameters, and response), and triggers Sentinel circuit breakers in case of exceptions (such as switching to a backup MCP Server instance). The frontend displays the execution progress in real time (e.g., progress bar, step status indicators) and supports pause / termination operations (implemented via the XXL-JOB API).

[0070] In some embodiments, the gateway service module includes: An ecosystem adaptation unit is used to synchronize the metadata of the MCPServer from at least one third-party MCP ecosystem platform through a scheduled task, and uniformly register it in the resource library for the visualization workflow orchestration module to orchestrate. The load balancing routing unit is used to route a request to a certain MCP Server to one of the healthy service instances according to a preset load balancing strategy when a call request to a certain MCP Server is received. If the MCP Server has multiple service instances deployed, the load balancing strategy includes a least connections strategy or a round-robin strategy.

[0071] It receives calls from clients to managed MCP Servers, parses the service name, interface parameters, and authentication information in the request, and determines whether the requested MCP Server exists in the resource repository; if not, it returns a "service not found" error response. Through the Nacos service discovery component, it queries a list of all service instances corresponding to the requested MCP Server and filters out instances with a healthy status.

[0072] The default algorithm uses the least connections algorithm, but users can manually switch to round-robin or weighted algorithms in the gateway configuration interface. It counts the current active connections of each healthy instance and selects the instance with the fewest active connections as the routing target. Requests are routed to each instance sequentially according to the instance list, ensuring even distribution. The probability of request allocation is calculated based on the weight value of each instance; instances with higher weights are more likely to be selected. The parsed interface parameters and authentication information are encapsulated into a request format conforming to the MCP protocol and forwarded to the selected service instance. If forwarding fails (e.g., an instance suddenly goes offline), the system automatically selects the next healthy instance for forwarding, with a maximum of three retries. If all three retries fail, an error response indicating that the service is temporarily unavailable is returned.

[0073] In this embodiment of the invention, the gateway service module further includes: The high availability guarantee unit is used to periodically send heartbeat detection requests to the service instances of the registered MCP Server. If no response is received from a service instance for a preset number of consecutive times, it is marked as unhealthy and the unhealthy service instance is temporarily excluded during routing until it recovers to health.

[0074] The gateway service module is configured to perform a unified security check on the request before routing the request. The check includes: the validity of the API Key; and / or whether the access frequency of the IP address initiating the request is within the allowed range.

[0075] The system provided in this application supports both manual and automatic shutdown (e.g., no calls for 24 hours). The shutdown process terminates containers or stops local processes via the Kubernetes API, while simultaneously updating the database status. It stores historical version code packages and metadata, supporting one-click rollback. Version status is marked using Nacos (e.g., "Currently Effective," "Historical Version"). Prometheus is used to monitor service health, triggering alerts when services encounter anomalies. Call counts, response times, error rates, etc., are recorded and visualized through a front-end dashboard, with data stored in a real-time database.

[0076] like Figure 2 As shown, this embodiment of the invention also provides an MCP integration management method that supports multi-task collaboration, including: MCP service management steps: Receive the MCP Server code package or remote service address submitted by the user, verify the protocol compatibility of the MCP Server, and after the verification is passed, deploy one or more service instances according to the communication mode of the MCP Server using the corresponding deployment strategy, and register the metadata of the MCP Server and the service instance information in the resource library to form an MCP Server list, so as to complete the management of the MCP Server. Collaboration space management steps: In response to user operations, create personal collaboration spaces and team collaboration spaces, and configure role permissions for members of the team collaboration spaces; Workflow orchestration and execution steps: Within the collaboration space, a visual drag-and-drop interface is provided to receive drag-and-drop connection operations performed by users based on the MCP Servers in the MCP Server list, so as to generate a workflow definition containing task execution logic; according to the workflow definition, the corresponding workflow instance is created and executed, and the execution status is fed back in real time. Gateway service steps: Receive call requests to managed MCP Servers, perform load balancing routing and protocol conversion to achieve seamless invocation of third-party MCP services; The workflow definitions orchestrated in the collaboration space can directly call the managed MCP Server when executed.

[0077] In some embodiments, the MCP service management steps include: Parse the code package submitted by the user to obtain the functional description, dependent environment, and interface parameter metadata of the MCP Server; The protocol compatibility of the MCP Server is verified based on the MCP protocol standard. Specifically, for SSE communication mode, it is verified whether the MCP Server exposes the specified SSE path and returns data of type text / event-stream; for STDIO communication mode, it is verified whether the MCP Server supports standard input / output pipe communication. The deployment strategy is adopted according to the communication mode of the MCP Server; for STDIO mode, the service process is started in the local sandbox environment; for SSE mode, the code package is built into a container image and deployed in the container orchestration platform. The metadata and service instance information of the successfully verified and deployed MCP Servers are stored in the database, and an index is created in the cache to form a list of MCP Servers in the resource library.

[0078] In some embodiments, the collaborative space management steps include: Create a personal collaboration space that is only visible and operable to the space creator; and create a team collaboration space that supports adding multiple collaboration members via invitation. Configure role permissions for members of the team collaboration space, the role permissions including: The administrator role grants full management access to the team space. The editor role has the authority to create and edit task workflows within the team space; The viewer role only has permission to view the task flow and results within the team space.

[0079] In some embodiments, the collaborative space management step further includes: When multiple users simultaneously edit the same task flow in the same team collaboration space, an operation conversion algorithm is used to resolve editing conflicts. When the role permissions of members in the team space change, the changed permission information will be updated to the cache in real time, and the session permissions of all currently logged-in users will take effect immediately. The system uses a WebSocket connection to push real-time state change events within the team collaboration space to the front-end interface.

[0080] In some embodiments, the workflow orchestration and execution steps include: User operations are received through the visual drag-and-drop interface to define at least one of the following task execution logics: serial orchestration, parallel orchestration, or conditional orchestration. Transform user drag-and-drop connection operations into structured workflow definition files; The workflow definition file is parsed and transformed into a standardized BPMN model, and the workflow instance is created and executed by the task scheduling engine.

[0081] In some embodiments, the workflow orchestration and execution steps further include: During the execution of the workflow instance, execution logs and performance metrics are recorded; In response to user commands or abnormal situations, the executing workflow instance is paused, terminated, or circuit-broken; wherein, the circuit-broken operation includes: when a call to a certain MCP Server fails consecutively, the request is automatically switched to a backup service instance of that MCP Server or another MCP Server that can perform the same function.

[0082] In some embodiments, the gateway service steps include: Metadata of the MCP Server is synchronized from at least one third-party MCP ecosystem platform via scheduled tasks and uniformly registered to the resource library; When a request to call a certain MCP Server is received, if the MCP Server has multiple service instances deployed, the request will be routed to one of the healthy service instances according to the least connection strategy or the round-robin strategy.

[0083] In some embodiments, the gateway service step further includes: Periodically send heartbeat requests to the registered MCP Server service instances. If no response is received for a preset number of consecutive times, mark them as unhealthy and exclude them during routing. Before routing the request, a unified security check is performed on the request. The security check includes verifying the validity of the API Key and / or verifying the access frequency of the IP address that initiated the request.

[0084] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An MCP integrated management system supporting multi-task cooperation, characterized by, Comprise: The MCP management module is configured to receive a user-submitted MCP Server code package or remote service address, check the protocol compatibility of the MCP Server, and, after the check passes, complete the deployment of one or more service instances according to a corresponding deployment strategy based on the communication mode of the MCP Server, and register the metadata and service instance information of the MCP Server in a resource library to form an MCP Server list, thereby completing the management of the MCP Server; The multi-task collaboration space creation module is configured to create a personal collaboration space and a team collaboration space in response to a user operation, and configure role permissions for members of the team collaboration space; The visual workflow orchestration module is configured to provide a visual drag-and-drop interface within the collaboration space, receive a user's drag-and-connect operation based on an MCP Server in the MCP Server list, generate a workflow definition containing task execution logic, and create and execute a corresponding workflow instance according to the workflow definition, and provide real-time feedback on the execution status; The gateway service module is configured to receive a call request for a managed MCP Server, perform load balancing routing and protocol conversion, and implement seamless calls to third-party MCP services. The visual workflow orchestration module can access the resource library to obtain the MCP Server list, so that the workflow definition generated based on the MCP Server list can directly call the managed MCP Server through the gateway service module when executed.

2. The MCP integrated management system supporting multi-task collaboration according to claim 1, wherein, The MCP management module comprises: A code analysis unit configured to analyze a user-submitted code package to obtain the function description, dependent environment, and interface parameter metadata of the MCP Server; A protocol checking unit configured to check the protocol compatibility of the MCP Server based on the MCP protocol standard, wherein, for the SSE communication mode, the protocol checking unit is configured to check whether the MCP Server exposes a specified SSE path and returns data of the text / event-stream type; and for the STDIO communication mode, the protocol checking unit is configured to check whether the MCP Server supports standard input / output pipe communication; A deployment execution unit configured to adopt a corresponding deployment strategy according to the communication mode of the MCP Server, wherein, for the STDIO communication mode, the deployment execution unit is configured to start a service process in a local sandbox environment; and for the SSE communication mode, the deployment execution unit is configured to build the code package into a container image and deploy it in a container orchestration platform; A registration indexing unit configured to store the metadata and service instance information of the MCP Server that has passed the check and deployment in a database, and establish an index in the cache according to the function classification and communication mode, thereby forming the MCP Server list in the resource library.

3. The MCP integrated management system supporting multi-task collaboration according to claim 2, wherein, The multi-task collaboration space creation module is configured to: Creating a personal collaboration space, which is only visible and operable to the space creator himself; and creating a team collaboration space, which supports adding multiple collaboration members by invitation; when detecting that multiple users are simultaneously editing the same task flow in the same team collaboration space, using an operation conversion algorithm to solve the editing conflict and realize real-time collaborative editing; Wherein, configuring role permissions for members of the team collaboration space includes: Administrators have complete management permissions for the team space, including managing members and permissions; Editors have the permission to create and edit task flows in the team space; Viewers only have the permission to view task flows and results in the team space.

4. The MCP integrated management system supporting multi-task collaboration according to claim 3, wherein, The multi-task collaboration space creation module further includes: A permission synchronization unit for updating the changed permission information to the cache in real time when the role permissions of the members in the team space are changed, and making the session permissions of all currently logged-in users take effect immediately; A state notification unit for pushing state change events in the team collaboration space to the front-end interface in real time through a WebSocket connection; the state change events include new task creation, task state update, or member change.

5. The MCP integrated management system supporting multitask collaboration according to claim 4, wherein, The visual drag-and-drop interface is configured to support users defining at least one of the following task execution logics through drag-and-drop and connection operations: Serial arrangement for defining task sequential execution relationship; Parallel arrangement for defining multiple tasks that can be executed simultaneously; Conditional arrangement for deciding the execution branch based on the execution result of the previous task.

6. The MCP integrated management system supporting multitask collaboration according to claim 5, wherein, The visual workflow arrangement module includes: A definition generation unit for converting the user's drag-and-drop connection operation into a structured workflow definition file, which defines the calling order of the MCP Server, parameter mapping relationship, and condition rules; A process engine unit for receiving the workflow definition file, parsing and converting it into a standardized BPMN model, and submitting it to a task scheduling engine to create and execute the workflow instance; An execution monitoring unit for recording execution logs and performance indicators during the execution of the workflow instance; and responding to user instructions or abnormal situations to pause, terminate, or fuse the executing workflow instance; wherein, the fuse operation includes: when consecutive failures occur in calling a certain MCP Server, automatically switching the request to a backup instance of the MCP Server or other MCP Server that can complete the same function.

7. The MCP integrated management system supporting multitask collaboration according to claim 6, wherein, The gateway service module includes: An ecological adaptation unit for synchronizing the metadata of the MCP Server from at least one third-party MCP ecological platform through a timing task, and uniformly registering it in the resource library for the visual workflow arrangement module to arrange; A load balancing routing unit for routing the request to one of the healthy service instances according to the preset load balancing strategy when receiving a call request for a certain MCP Server, if the MCP Server is deployed with multiple service instances; the load balancing strategy includes the least connection number strategy or the round-robin strategy.

8. The MCP integrated management system supporting multitask collaboration according to claim 7, wherein, The gateway service module further includes: A high-availability assurance unit is configured to periodically send heartbeat detection requests to the registered service instances of the MCP Server, and if no response is received from a service instance for a preset number of times, the service instance is marked as unhealthy and temporarily excluded from routing until it recovers.

9. The MCP integrated management system supporting multitask collaboration according to claim 8, wherein, The gateway service module is configured to perform unified security verification on the request before routing the invocation request, and the verification content includes: validity of the APIKey; and / or whether the access frequency of the IP address initiating the request is within the allowed range.

10. A method for integrated management of MCPs supporting multi-task collaboration, characterized by, The method comprises the following steps: Receiving a user-submitted MCP Server code package or remote service address, checking the protocol compatibility of the MCP Server, and after passing the check, deploying one or more service instances according to the corresponding deployment strategy based on the communication mode of the MCP Server, and registering the metadata and service instance information of the MCP Server in the resource library to form a MCP Server list, thereby completing the management of the MCP Server; In response to a user operation, creating a personal collaboration space and a team collaboration space, and configuring role permissions for members of the team collaboration space; In the collaboration space, a visual drag-and-drop interface is provided to receive a user's drag-and-connect operation based on the MCP Server in the MCP Server list to generate a workflow definition containing task execution logic; according to the workflow definition, a corresponding workflow instance is created and executed, and the execution status is fed back in real time; Receiving an invocation request for the managed MCP Server, performing load balancing routing and protocol conversion to achieve seamless invocation of the third-party MCP service; The workflow definition generated in the collaboration space can directly invoke the managed MCP Server when executed.

Images