Bank card manager method and bank card manager system based on agent

Abstract

The present application relates to an agent-based bank card manager system and method thereof. The system comprises: a user interaction layer receiving a natural language request; a large model service layer for providing a large language model; a business service layer containing MCP service nodes for executing MCP services; a service registration and calling layer for executing registration and query of MCP service nodes and the following business agents; an agent system comprising at least general agents and business agents, wherein the general agents obtain the natural language request from the user interaction layer and parse the natural language request to obtain a request intention and a request element by calling the large language model, the general agents match and call corresponding business agents based on the request intention by querying the service registration and calling layer, and the business agents call MCP service nodes of the business service layer based on the request element to execute MCP services by querying the service registration and calling layer.

Description

Technical Field

[0001] This invention relates to the field of financial payment, and more specifically, to a bank card management method and system based on intelligent agents. Background Technology

[0002] Traditional banking applications and third-party tools use a menu-driven interface, requiring users to switch between multiple pages to complete operations such as repayment and transfer. The functional paths are fragmented and lack guidance. For example, users need to manually link steps such as bill inquiry and repayment, which presents difficulties for the elderly and limits their acceptance and frequency of use.

[0003] Furthermore, existing intelligent agents only support single-command queries (such as balance or bill), and cannot perform transaction-related operations or automatically plan tasks based on user goals. For example, when a user requests a combined request such as "automatically pay the minimum amount next week and pay the remaining amount in installments," the system cannot break down the task and execute it, requiring manual intervention.

[0004] Furthermore, the traditional MCP protocol relies on point-to-point communication, requiring pre-configuration of the IP addresses of all target servers and lacking service registration and discovery mechanisms. This results in weak system disaster recovery capabilities, difficulty in dynamic scaling, inability to adapt to load changes in multi-agent collaborative scenarios, and limitations on rapid business expansion.

[0005] In summary, existing technologies have shortcomings in terms of interaction efficiency, task intelligence, and system architecture, making it difficult to meet users' needs for unified management of multiple cards across banks, natural language interaction, and high scalability. Summary of the Invention

[0006] To address the problems in the prior art, this invention aims to provide a bank card management system based on an intelligent agent, comprising: The user interaction layer is used to receive natural language requests; The large model service layer is used to provide large language models for parsing natural language. The business service layer contains one or more MCP service nodes for performing MCP services; The service registration and invocation layer is used to perform the registration and query of the one or more MCP service nodes and the one or more business intelligent agents described below; The intelligent agent system includes at least a general intelligent agent that performs routing functions and one or more business intelligent agents that perform bank card management functions. The general intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent. The business intelligent agent matches and invokes one or more MCP service nodes of the business service layer based on the request elements to execute one or more MCP services.

[0007] Optionally, the service registration and invocation layer includes a registry center for registering and querying MCP services and business intelligent agents. The registration center pre-registers agent semantic vectors generated from business capability descriptions of business agents and MCP service semantic vectors generated from description texts of MCP services based on MCP service nodes.

[0008] Optionally, the intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the registration center of the service registration and invocation layer to invoke a business intelligent agent with matching business capabilities and establishes a connection with the invoked business intelligent agent. The general-purpose intelligent agent synchronizes the requested elements to the business intelligent agent that established the connection; and The business intelligence agent that establishes the connection invokes one or more MCP service nodes that match the service capabilities by querying the registration center of the service registration and invocation layer based on the request elements, in order to execute one or more MCP services.

[0009] Optionally, the general intelligent agent, based on the request intent, invokes a business intelligent agent matching the business capabilities by querying the registry center of the service registration and invocation layer, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the agent registered in the registration center of the service registration and invocation layer.

[0010] Optionally, the business intelligence agent establishing the connection, based on the request elements, queries the service registration and invocation layer's registry center to invoke one or more MCP service nodes matching the service capabilities, including: The business intelligence agent that establishes the connection invokes one or more corresponding MCP service nodes based on the similarity between the semantic vector of the request element and the semantic vector of the MCP service registered in the registration center of the service registration and invocation layer.

[0011] Optionally, the intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

[0012] Optionally, the intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

[0013] Optionally, the business intelligence agent includes: A smart card manager agent used to implement bank card management functions; and / or One or more assistive agents used to implement assistive functions.

[0014] Optionally, when the business intelligence agent includes a card management intelligence agent and an auxiliary intelligence agent, The intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. The general-purpose agent synchronizes the requested elements to the card manager agent and the auxiliary agent that established the connection; and The card manager intelligent agent and the auxiliary intelligent agent that establish the connection query the service registration and invocation layer based on the request elements to invoke multiple MCP service nodes of the business service layer to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of multiple MCP services into natural language execution results using the large language model, and the large language model then summarizes these results and sends them to the card manager intelligent agent that establishes the connection; and The card manager intelligent agent that establishes the connection provides the aggregated execution results to the user interaction layer through the general intelligent agent.

[0015] Optionally, the registration center has the following features: The receiving module is configured to receive in advance the registration requests of the one or more MCP service nodes and the one or more business intelligent agents, wherein the registration request includes an MCP service semantic vector generated based on the description text of the MCP service of the MCP service node and an intelligent agent semantic vector generated based on the business capability description of the business intelligent agent. A database is used to store the semantic vectors of the MCP service and the semantic vectors of the agent; The calculation module is used to calculate the first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and to filter out business agents with a first similarity higher than a first threshold as matching results; and to calculate the second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and to filter out MCP service nodes with a second similarity higher than a second threshold as matching results.

[0016] Optionally, the MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. The MCP service node queries the registration center for neighboring MCP service nodes with a semantic vector similarity higher than a third threshold; and MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

[0017] Optionally, the MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

[0018] This invention discloses a bank card management method based on intelligent agents, which utilizes a user interaction layer, a large model service layer, a business service layer, a service registration and invocation layer, and an intelligent agent system. The intelligent agent system includes a general intelligent agent and one or more business intelligent agents. The large model service layer provides a large language model capable of parsing natural language. The business layer includes one or more MCP service nodes for executing MCP services. The bank card management method based on intelligent agents includes the following steps: The user interaction layer is used to receive natural language requests from users; The general-purpose intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general-purpose intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent; and The business intelligence agent queries the service registration and invocation layer based on the request elements to invoke one or more MCP service nodes of the business service layer to execute one or more MCP services.

[0019] Optionally, the general intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent, including: The general agent, based on the request intent, queries the service registration and invocation layer to invoke a business agent with matching business capabilities and establishes a connection with the invoked business agent.

[0020] Optionally, it further includes: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

[0021] Optionally, it further includes: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

[0022] Optionally, the general intelligent agent, based on the request intent, invokes a business intelligent agent matching the business capabilities by querying the service registration and invocation layer, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the business capabilities of the business agents already registered in the service registration and invocation layer.

[0023] Optionally, the business intelligent agent includes: a card manager intelligent agent for implementing bank card management functions; and one or more auxiliary intelligent agents for implementing auxiliary functions. Based on the request intent, the general intelligent agent queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. Furthermore, the general intelligent agent synchronizes the request elements to the established card manager intelligent agent and auxiliary intelligent agent. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection call each invoke multiple MCP service nodes of the business service layer by querying the service registration and invocation layer based on the request elements to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of the multiple MCP services into natural language execution results using the large language model, and the large language model summarizes them to the card manager intelligent agent that establishes the connection. The card manager intelligent agent that establishes the connection provides the summarized execution results to the user interaction layer through the general intelligent agent.

[0024] Optionally, the service registration and invocation layer has a registry center for registering and querying MCP services and business agents.

[0025] The registry center is configured to perform the following actions: The system receives registration requests from one or more MCP service nodes and one or more business agents in advance. The registration requests include an MCP service semantic vector generated from the description text of the MCP service of the MCP service node and an agent semantic vector generated from the description of the business capabilities of the business agent. Store the MCP service semantic vector and the agent semantic vector; The system calculates a first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and selects business agents with a first similarity higher than a first threshold as matching results. It also calculates a second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and selects MCP service nodes with a second similarity higher than a second threshold as matching results.

[0026] Optionally, the MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. The MCP service node queries the registration center for neighboring MCP service nodes with a semantic vector similarity higher than a third threshold; and MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

[0027] Optionally, the MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

[0028] The present invention provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned agent-based bank card management method.

[0029] A computer device according to one aspect of the present invention includes a storage module, a processor, and a computer program stored on the storage module and executable on the processor, wherein the processor executes the computer program to implement the aforementioned agent-based bank card management method.

[0030] A computer program product according to one aspect of the present invention includes a computer program that, when executed by a processor, implements the aforementioned agent-based bank card management method. Attached Figure Description

[0031] The described and other objects and advantages of the invention will become more fully clear from the following detailed description taken in conjunction with the accompanying drawings, wherein like or similar elements are denoted by the same reference numerals.

[0032] Figure 1 This is a schematic diagram of the structure of a bank card management system based on an intelligent agent according to an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram illustrating the use of the intelligent agent-based bank card management system of the present invention to achieve complex task orchestration.

[0034] Figure 3 This is a flowchart illustrating a specific process for implementing complex task orchestration using the agent-based bank card management system of the present invention.

[0035] Figure 4 This is a schematic diagram illustrating the existing MCP workflow.

[0036] Figure 5 This is a schematic diagram illustrating the MCP service discovery aggregation mechanism based on semantic vector similarity of the present invention. Detailed Implementation

[0037] The following are some embodiments of the present invention, intended to provide a basic understanding of the invention. They are not intended to identify key or decisive elements of the invention or to limit the scope of protection sought.

[0038] For purposes of brevity and illustrativeness, this document primarily refers to exemplary embodiments thereof to describe the principles of the invention. However, those skilled in the art will readily recognize that the same principles are equivalently applicable to all types of agent-based bank card management methods and systems, and that these same principles can be implemented therein, with any such variations not departing from the true spirit and scope of the invention.

[0039] Furthermore, reference is made in the following description to the accompanying drawings, which illustrate specific exemplary embodiments. Electrical, mechanical, logical, and structural modifications may be made to these embodiments without departing from the spirit and scope of the invention. Moreover, while features of the invention are disclosed in conjunction with only one of several embodiments, this feature may be combined with one or more other features of other embodiments if desired and / or advantageous for any given or identifiable function. Therefore, the following description should not be considered limiting in any sense, and the scope of the invention is defined by the appended claims and their equivalents.

[0040] Terms such as “possessing” and “comprising” indicate that, in addition to having units (modules) and steps that are directly and explicitly stated in the specification and claims, the technical solution of the present invention does not exclude the presence of other units (modules) and steps that are not directly or explicitly stated.

[0041] The bank card management system based on intelligent agents of the present invention includes: The user interaction layer is used to receive natural language requests from users. The large model service layer is used to provide large language models capable of parsing natural language. The business service layer contains one or more MCP service nodes for performing MCP services; The service registration and invocation layer is used to perform the registration and query of the one or more MCP service nodes and the one or more business intelligent agents described below; The intelligent agent system includes at least a general intelligent agent and one or more business intelligent agents. The general intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent. The business intelligent agent invokes one or more MCP service nodes of the business service layer based on the request elements to execute one or more MCP services.

[0042] According to the present invention, the bank card management system based on intelligent agents receives natural language requests through the user interaction layer, parses the request intent and request elements using the large model service layer, and the intelligent agent system matches and calls the business intelligent agent according to the parsing results. The business intelligent agent then calls the MCP service node of the business service layer to execute specific services. This enables the intelligent and unified management of bank cards, improves operational efficiency and user experience, and solves the problems of cumbersome operation and lack of unified entry point in traditional bank APPs.

[0043] The intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the service registration and invocation layer to invoke a business intelligent agent whose business capabilities match and establishes a connection with the invoked business intelligent agent. The general-purpose intelligent agent synchronizes the requested elements to the business intelligent agent that established the connection; and The business intelligence agent that establishes the connection queries the service registration and invocation layer based on the request elements to invoke one or more MCP service nodes of the business service layer to execute one or more MCP services.

[0044] Therefore, the general intelligent agent parses natural language requests through a large language model, queries the service registration and invocation layer based on the request intent, calls and connects to business intelligent agents that match the business capabilities, and synchronizes request elements. This configuration enables precise processing of requests and dynamic invocation of business intelligent agents, improving the system's flexibility and response speed.

[0045] The intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

[0046] Therefore, the business intelligence agent that establishes the connection converts the execution results or summary results of the MCP service into natural language using a large language model, making it easier for users to understand. This function enhances the user-friendliness of the system and makes the execution results more intuitive and easy to understand.

[0047] The intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

[0048] Thus, the business intelligence agent that establishes the connection provides the execution results of natural language to the user interaction layer through the general intelligence agent, realizing the instant feedback and display of the execution results. This process ensures that users can obtain the service execution status in a timely manner, improving the interactivity and transparency of the system.

[0049] The general intelligent agent, based on the request intent, invokes a business intelligent agent matching the business capabilities by querying the service registration and invocation layer, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the business capabilities of the business agents already registered in the service registration and invocation layer.

[0050] Therefore, the general intelligent agent can invoke the corresponding business intelligent agent based on the similarity between the semantic vector of the request intent and the semantic vector of the business capability of the registered business intelligent agent. This can improve the accuracy and efficiency of business intelligent agent invocation, realize semantic-based intelligent matching, and achieve high matching accuracy and fast matching speed.

[0051] The business intelligence agent includes: A smart card manager agent used to implement bank card management functions; and / or One or more assistive agents used to implement assistive functions.

[0052] Therefore, when the business intelligence agent includes the card manager intelligence agent and the auxiliary intelligence agent, the system can parse natural language requests, call and connect to the matching business intelligence agent, synchronize request elements, execute multiple MCP services, and convert the execution results into natural language and provide them to the user. This enables automated processing of complex requests and intuitive display of results, improving the system's intelligence level and user experience.

[0053] In the case where the business intelligence agent includes a card management intelligence agent and an auxiliary intelligence agent... The intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. The general-purpose agent synchronizes the requested elements to the card manager agent and the auxiliary agent that established the connection; and The card manager intelligent agent and the auxiliary intelligent agent that establish the connection query the service registration and invocation layer based on the request elements to invoke multiple MCP service nodes of the business service layer to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of multiple MCP services into natural language execution results using the large language model, and the large language model then summarizes these results and sends them to the card manager intelligent agent that establishes the connection; and The card manager intelligent agent that establishes the connection provides the aggregated execution results to the user interaction layer through the general intelligent agent.

[0054] Therefore, the service registration and invocation layer has a registry center for registering and querying MCP services and business intelligence agents, realizing unified management and dynamic discovery of services, improving the scalability and flexibility of the system, and facilitating the rapid expansion and flexible deployment of subsequent business.

[0055] The service registration and invocation layer includes a registration center for registering and querying MCP services and business intelligent agents.

[0056] The registration center has the following features: The receiving module is configured to receive in advance the registration requests of the one or more MCP service nodes and the one or more business intelligent agents, wherein the registration request includes an MCP service semantic vector generated based on the description text of the MCP service of the MCP service node and an intelligent agent semantic vector generated based on the business capability description of the business intelligent agent. A database is used to store the semantic vectors of the MCP service and the semantic vectors of the agent; The calculation module is used to calculate the first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and to filter out business agents with a first similarity higher than a first threshold as matching results; and to calculate the second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and to filter out MCP service nodes with a second similarity higher than a second threshold as matching results.

[0057] Therefore, the registration center receives registration requests from MCP service nodes and business agents in advance, stores semantic vectors, calculates the similarity between the request intent and the registered vectors, and filters out the matching results, realizing semantic-based intelligent service discovery and matching, and improving the accuracy and efficiency of service calls.

[0058] The MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. The MCP service node queries the registration center for neighboring MCP service nodes with a semantic vector similarity higher than a third threshold; and MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

[0059] Thus, MCP service nodes generate semantic vectors and register them with the registry center, query and connect neighboring nodes with high similarity, and aggregate them into clusters. This enables automatic discovery and aggregation of MCP service nodes, improves the system's disaster recovery capability and scalability, and reduces latency and communication overhead.

[0060] The MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

[0061] Therefore, each MCP service node maintains semantic vector similarity with its neighboring nodes. When the similarity is lower than the threshold, the aggregated cluster is disbanded and re-aggregation is triggered, which can ensure the dynamic adjustment and optimization of service clusters and maintain the efficiency and stability of the system.

[0062] The bank card management method based on intelligent agents of the present invention is implemented using a user interaction layer, a large model service layer, a business service layer, a service registration and invocation layer, and an intelligent agent system. The intelligent agent system includes a general intelligent agent and one or more business intelligent agents. The large model service layer provides a large language model capable of parsing natural language. The business service layer includes one or more MCP service nodes for performing MCP services. The bank card management method based on intelligent agents of the present invention includes the following steps: The user interaction layer is used to receive natural language requests from users; The general-purpose intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general-purpose intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent; and The business intelligence agent queries the service registration and invocation layer based on the request elements to invoke one or more MCP service nodes of the business service layer to execute one or more MCP services.

[0063] The general intelligent agent, by querying the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent, includes: The general agent, based on the request intent, queries the service registration and invocation layer to invoke a business agent with matching business capabilities and establishes a connection with the invoked business agent.

[0064] The bank card management method based on intelligent agents of the present invention further includes: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

[0065] The bank card management method based on intelligent agents of the present invention further includes: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

[0066] The general intelligent agent, based on the request intent, invokes a business intelligent agent matching the business capabilities by querying the service registration and invocation layer, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the business capabilities of the business agents already registered in the service registration and invocation layer.

[0067] The business intelligent agents include: a card manager intelligent agent for implementing bank card management functions; and one or more auxiliary intelligent agents for implementing auxiliary functions. Based on the request intent, the general intelligent agent queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. Furthermore, the general intelligent agent synchronizes the request elements to the established card manager intelligent agent and auxiliary intelligent agent. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection call each invoke multiple MCP service nodes of the business service layer by querying the service registration and invocation layer based on the request elements to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of the multiple MCP services into natural language execution results using the large language model, and the large language model summarizes them to the card manager intelligent agent that establishes the connection. The card manager intelligent agent that establishes the connection provides the summarized execution results to the user interaction layer through the general intelligent agent.

[0068] The service registration and invocation layer includes a registration center for registering and querying MCP services and business intelligent agents.

[0069] The registry center is configured to perform the following actions: The system receives registration requests from one or more MCP service nodes and one or more business agents in advance. The registration requests include an MCP service semantic vector generated from the description text of the MCP service of the MCP service node and an agent semantic vector generated from the description of the business capabilities of the business agent. Store the MCP service semantic vector and the agent semantic vector; The system calculates a first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and selects business agents with a first similarity higher than a first threshold as matching results. It also calculates a second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and selects MCP service nodes with a second similarity higher than a second threshold as matching results.

[0070] The MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. The MCP service node queries the registration center for neighboring MCP service nodes with a semantic vector similarity higher than a third threshold; and MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

[0071] The MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

[0072] According to this invention, the system can support the management of all bank cards under a user's name from all banks, unify all bank card management functions in the form of a dialogue agent, and support complex task flow orchestration, enabling users to perform full-process card management operations on a single page. Furthermore, this invention proposes a method for MCP service registration and discovery that integrates centralized indexing capabilities with a decentralized semantic aggregation mechanism. This method can achieve automatic discovery, aggregation, updating, and management of semantic relationships between agents in a multi-agent environment, realizing "self-registration and self-discovery" of the agent network. Further, this invention proposes a framework for pluggable card management agents based on artificial intelligence and MCP services. This agent framework supports rapid plugging and unplugging of functions and efficient collaboration between agents, enabling rapid access to different business scenarios, thereby improving the overall intelligence level and horizontal scalability of the system.

[0073] Next, a specific embodiment of the intelligent agent-based bank card management system of the present invention will be described.

[0074] Figure 1 This is a schematic diagram of the structure of a bank card management system based on an intelligent agent according to an embodiment of the present invention. Figure 1 As shown, the bank card management system based on intelligent agents of the present invention constructs a flexible, intelligent and professional AI card management system through hierarchical design and collaboration with intelligent agents.

[0075] like Figure 1 As shown, the bank card management system based on intelligent agents according to one embodiment of the present invention mainly consists of five core parts: (1) User interaction layer 100 Users access the system via mobile devices. Figure 1 The user's mobile phone represents the user interaction layer, serving as the starting point for service access. All card management functions (repayment, transfer, etc.) and their results feedback are completed through interaction with this terminal.

[0076] (2) Intelligent Agent System 200 The intelligent agent system 200 includes: The general-purpose intelligent agent 210 is responsible for the initial processing and distribution of user requests. It integrates routing functions to ensure accurate request guidance and provides general question-and-answer functions (plug-in type) such as card details, repayment, transfer, fingerprint analysis, and general knowledge in a plug-in format to quickly respond to users' basic query and operation needs. The process orchestration agent 220 serves as the "command center" for collaboration among agents, connecting general agents with other business agents, streamlining business process logic, and ensuring the orderly execution of services across multiple scenarios. Card Manager Intelligent Agent 230 focuses on in-depth processing of financial scenarios, such as repayment scenario processing, transfer scenario processing, card details scenario processing, and income and expenditure analysis scenario processing. For different financial business needs, it calls the underlying MCP (Model Context Protocol) service to achieve scenario-based and professional business processing.

[0077] In addition, the general intelligent agent 210, the process orchestration intelligent agent 220, and the card manager intelligent agent 230 communicate with the AI ​​gateway via HTTP.

[0078] (3) Service registration and invocation layer 300 The service registration and invocation layer 300 may include the MCP service registration center and the business intelligent agent registration center (hereinafter referred to as the "registration center"). The MCP service registration center and the business intelligent agent registration center are the core of service governance, responsible for the registration and query of intelligent agents and MCP services, ensuring that services are discoverable and manageable. Among them, the MCP client acts as a client agent, connecting the card manager intelligent agent and the MCP service, uniformly encapsulating the service invocation logic, and simplifying the interaction complexity.

[0079] (4) Business service layer 400 MCP services can be broken down according to financial business types, such as repayment services, transfer services, card management services, order services, and other services. Each service is deployed independently, and through the "service + resource tools" model, it provides standardized and professional business capability support for upper-layer intelligent agents.

[0080] (5) Large model service layer 500 By leveraging the AI ​​gateway and connecting with LLM (Large Language Model) and intelligent agent platforms (such as HiAgent), the system is infused with natural language understanding and intelligent decision-making capabilities, thereby enhancing the intelligence level of the card management service and realizing the integrated application of financial scenarios and large model capabilities.

[0081] Figure 2 This is a schematic diagram illustrating the use of the intelligent agent-based bank card management system of the present invention to achieve complex task orchestration.

[0082] For example, a task might be, "How many payments do I have this month? Is my balance sufficient?", then... Figure 2 The diagram in the middle illustrates the relevant process to achieve this task, which mainly includes the following four steps: A: Check Card List: Call the MCP service center to obtain the user card list service. This step aims to obtain the card list information owned by the user through the MCP service, providing basic data for subsequent card queries. B: Check credit card statements: Based on the results of the card list, check the statements of the credit cards in the card list. This step is to further initiate statement query requests for the credit cards in the list after obtaining the user's card list in order to obtain the consumption statement information of each credit card. C: Check Debit Card Balance: Also based on the card list results, this step checks the balance of debit cards listed in the card list. This step aims to further perform balance checks on the debit cards in the card list to obtain their fund balance information; and D: Application Calculation and Large Model Polishing: The application calculates the final answer based on the bill and balance information obtained from the previous query, and returns it to the user after being polished by the large model. This step aims to integrate and calculate credit card bill data and debit card balance data, which may include statistical and analytical operations to form the final information required by the user. Then, the language processing capabilities of the large model are used to polish and optimize the calculation results, making the information returned to the user more readable, accurate and user-friendly.

[0083] Figure 3 This is a flowchart illustrating a specific process for implementing complex task orchestration using the agent-based bank card management system of the present invention.

[0084] Figure 3 The problem to be solved (Q1) is: "Can my balance be used to pay off the credit card bill due before the 15th of this month?", which can be mainly divided into five stages from S1 to S5: (1) S1: MCP service registration Multiple MCP services are registered in the registration center, including: S1-1: Query the user card list service, for example, the registered name is qryCardList, and the parameter is userid; S1-2: Debit card balance query service, for example, registered as qryDebitCardBalance, with parameters userid and List. <debitcardno>; S1-3: Credit card bill inquiry service, for example, registered under the name qryCreditCardBill, with parameters userid and List. <creditcardno>.

[0085] (2) S2: Card Manager Intelligent Agent Initialization Service discovery and retrieval of available MCP services: The Card Manager agent obtains information on registered MCP services from the registry center for subsequent invocation.

[0086] (3) S3: Business intelligence agent registration.

[0087] (4) S4: General routing layer initialization Business agent discovery and MCP service discovery: pull available business agents and MCP services to prepare for routing subsequent requests.

[0088] (5) S5: User Request Phase The user request phase mainly includes the following processes: S5-1: Send user question Q1, which is the question the user asks the system: "Can my balance be used to pay off the credit card bill due before the 15th of this month?" S5-2: The LLM is sent through the general routing layer. The LLM parses the intent to call the business agent. Specifically, it parses the user's question intent, determines the CardManager business agent to be called, and the context will transmit the business agent's metadata (such as prompting that the available business agent is CardManager). S5-3: Returns structured parameters to specify the business intelligence agent and specific function to be called, for example: bizAgentNm=CardManager,function=BankCardManagementTool; S5-4: Call the Card Manager intelligent agent for processing; S5-5: LLM parsing intent, context transmission of service metadata, specifically: further parsing intent, context transmission of MCP service metadata (such as prompting that available services are qryCardList, qryDebitCardBalance, qryCreditCardBill). S5-6: Returns structured parameters, which include the service name, parameters, and call sequence, for example: [{"name":"qryCardList", "param":"userid", "callSequence":"1"}, {"name":"qryDebitCardBalance", "param":"userid&List"} <debitcardno>", "callSequence":"2"}, {"name":"qryCreditCardBill", "param":"userid&List <creditcardno>", "callSequence":"3"}]; S5-7: Send an MCP request to the target service qryCardList to obtain the list of user cards; S5-8: Returns a response in MCP format; S5-9: Send an MCP request to the target service qryDebitCardBalance to retrieve the debit card balance; S5-10: Send an MCP request to the target service qryCreditCardBill to retrieve the credit card bill; S5-11: Returns a response in MCP format; S5-12: Returns a response in MCP format; S5-13: Calculate whether the bill can be paid off based on the obtained debit card balance information and credit card bill information, and obtain the calculation result; S5-14: Provide the acquired information and settlement results to the LLM through the AI ​​gateway; S5-15: LLM converts the computation results into natural language descriptions based on contextual information through a large model, such as generating text: "Your credit card bill due before the 15th of this month totals 5237.00 yuan, and your current total debit card balance is 26000.00 yuan, which is sufficient to pay off the bill. The bill and balance information for each card is as follows: [...] Detailed information for each card]"; S5-16: Return the result to the general route; S5-17: Return the results to the user, completing the entire Q&A process.

[0089] In the steps described above, when a user inputs a request, such as "Can my balance be used to pay off my credit card bill due before the 15th of this month?" (i.e., S5-1), the general routing layer sends the user request to the LLM via the AI ​​gateway for core intent analysis and key information element extraction (i.e., S5-2). The LLM identifies the core intent of the user request and labels it with intent category tags such as "bill inquiry + balance verification." Based on the user input, it initially extracts request elements such as user identity and time period. The LLM returns the analysis results to the general routing layer (i.e., S5-3). The general routing layer receives the information returned by the LLM, queries the registry center for the agent with the most similar capabilities based on the intent category tags sent by the LLM, and establishes a connection. The general routing layer encapsulates the extracted key information into standard request parameters and synchronizes them to the card manager agent (i.e., S5-4). The Card Manager agent first performs intent classification based on the messages transmitted from the general routing layer, and checks whether the user's request matches the agent's preset workflow. If the intent classifier confidence is above 80%, the MCP service is directly called according to the preset workflow. If the intent classifier confidence is below 80%, specific intent analysis is performed through LLM (i.e., S5-5), and the appropriate MCP service is selected for processing.

[0090] If a user's request involves multiple business scenarios, such as "Please check my China Merchants Bank card balance to see if I can pay for the G1 high-speed train ticket departing the day after tomorrow," the user request clearly involves two business scenarios: bank card balance inquiry and train ticket inquiry. The intent categories are "bank card balance inquiry" and "train ticket inquiry." The Card Manager intelligent agent cannot respond to the user's request alone and needs the assistance of intelligent agents with relevant capabilities (e.g., travel intelligent agents, not illustrated here). The Card Manager intelligent agent searches the registry center to find the intelligent agent whose business scenario is closest to the user's second intent, establishes a connection with it, and transmits the context. Other intelligent agents call the corresponding MCP service to process the request and return the results to the Card Manager intelligent agent.

[0091] Finally, the Card Manager agent sends its own computation results, the computation results of other agents (if assistance from other agents is involved), and the dialogue context to the LLM (i.e., S5-14). The LLM aggregates and processes all the results and converts them into natural language, while also refining the returned natural language results (i.e., S5-15). The Card Manager agent then returns the natural language results generated by the LLM to the user via the general routing layer (i.e., S5-16, S5-17).

[0092] Next, the MCP service registration and discovery process described above will be explained. Before explaining the MCP service registration and discovery process of the present invention, the existing MCP workflow will be explained first.

[0093] Figure 4 This is a schematic diagram illustrating the existing MCP workflow.

[0094] like Figure 4 As shown, in the existing MCP protocol, the main steps of interaction between the client and the service are as follows: 1. User-initiated query: Users submit questions and initiate query requests through applications (such as Claude Desktop or Cursor); 2: MCP Client Obtains Tool List: The MCP client obtains a list of available tools and capabilities from the MCP service; 3: LLM Decision-Making: Based on the questions raised by users and the list of tools obtained, LLM decides whether to use tools and which tools to use. 4A: Tool Invocation: If the LLM decides to use a tool, the tool invocation is performed through the MCP service; 4B: Direct Response: If the LLM determines that no tool is needed, the response content is generated directly without the need for tool invocation steps; 5: Generate Response: Based on all the information obtained (including the results of tool calls, if any), LLM generates the final response content and sends the response back to the user.

[0095] The existing architecture of MCP described above leads to a high degree of coupling between the client and the service. Therefore, this invention proposes the following solution to address the problem of high coupling between the client and the service and difficulty in dynamic scaling.

[0096] Figure 5 This is a schematic diagram illustrating the MCP service discovery aggregation mechanism based on semantic vector similarity of the present invention.

[0097] like Figure 5 As shown, the MCP service registration and aggregation process includes: When an MCP service (such as MCP service A~D, MCP service D) goes online, it collects description information of its own tool, such as "query credit card bill information"; The descriptive text is fed into a semantic embedding model (such as OpenAI embedding) for vectorization to generate semantic vectors; The MCP service sends a registration request to the registry center, which stores the node metadata sent by the MCP service in the semantic vector database for subsequent queries. The metadata includes information such as IP address, port, semantic vector, and running status. Once an MCP service is successfully registered, it queries the registry center for a list of MCP services that are semantically similar to itself (e.g., similarity >= 0.8) and establishes connections with these services. Both parties establish a heartbeat mechanism and periodically exchange capability summaries and semantic vectors to achieve dynamic aggregation of MCP services; Each node maintains the average semantic similarity with its neighboring nodes. When the similarity is lower than a specified value (such as 0.8), it is considered that semantic drift has occurred, triggering a re-aggregation process. As service capabilities gradually stabilize, services with similar capabilities automatically aggregate, while dissimilar services naturally disperse, forming service clusters. These clusters then become logical intelligent agents. Figure 5 It can be represented as intelligent agent A being aggregated from MCP services A to D, and intelligent agent being composed of MCP service D alone.

[0098] As an example of service load balancing, clients can select instances from the list of available instances based on selection strategies (round-robin, random, weighted random, etc.) to distribute request pressure, avoid single points of congestion, and automatically switch to other healthy instances when a request fails or a response times out.

[0099] As described above, the agent-based bank card management system and method of the present invention enable full-process card management operations, including repayment, transfer, and bill analysis, to be performed on a single page using the card manager agent, without the need to switch between different pages, significantly improving operational efficiency and convenience. Moreover, the card manager agent can manage all bank cards under a user's name from all banks and unify all bank card management functions in the form of a conversational agent, solving the problem that traditional single-card or single-bank management methods cannot handle multi-card scenarios, thus enhancing the system's applicability and user experience.

[0100] Furthermore, the agent-based bank card management system and method of this invention propose an MCP service discovery and aggregation mechanism based on semantic vector similarity, and an MCP service registration and discovery method that integrates centralized indexing capabilities with a decentralized semantic aggregation mechanism. This enables automatic discovery, aggregation, updating, and management of semantic relationships between agents. The registry center stores all current MCP service metadata. For multi-agent collaboration and cross-agent invocation scenarios, the registry center can quickly query corresponding MCP services, reducing query latency. Based on this mechanism, the client can quickly match the most suitable service instance according to actual business needs, thereby improving the system's flexibility, reliability, and scalability in multi-agent collaboration and dynamic business scenarios. Thus, through the decentralized semantic aggregation mechanism, the MCP client no longer needs to configure a fixed MCP service IP; semantically similar MCP services can autonomously discover each other and establish chains, thereby reducing system coupling and improving the system's flexibility, reliability, and scalability in multi-agent collaboration and dynamic business scenarios.

[0101] The intelligent agent-based bank card management system and method of the present invention support the "understanding and expression" capabilities of AI, enabling fast and high-quality services to client users. For example, it processes user requests through natural language and provides rapid responses, which significantly improves user experience and promotes the transformation of financial business.

[0102] The intelligent agent-based bank card management system and method of the present invention can rapidly deploy new functional modules (such as other business intelligent agents) in different business scenarios, achieve seamless connection of business processes, significantly improve the overall intelligence level and horizontal expansion capability of the system, and provide an efficient and sustainable technical foundation for subsequent business function expansion and multi-scenario access.

[0103] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art can conceive of other feasible variations or substitutions based on the technical scope disclosed in the present invention, and such variations or substitutions are all covered within the scope of protection of the present invention. In the absence of conflict, the embodiments of the present invention and the features thereof can also be combined with each other. The scope of protection of the present invention is determined by the claims.< / creditcardno> < / debitcardno> < / creditcardno> < / debitcardno>

Claims

1. A bank card management system based on intelligent agents, characterized in that, include: The user interaction layer is used to receive natural language requests; The large model service layer is used to provide large language models for parsing natural language. The business service layer contains one or more MCP service nodes for performing MCP services; The service registration and invocation layer is used to perform the registration and query of the one or more MCP service nodes and the one or more business intelligent agents described below; The intelligent agent system includes at least a general intelligent agent that performs routing functions and one or more business intelligent agents that perform bank card management functions. The general intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent. The business intelligent agent matches and invokes one or more MCP service nodes of the business service layer based on the request elements to execute one or more MCP services.

2. The bank card management system based on intelligent agents as described in claim 1, characterized in that, The service registration and invocation layer has a registration center for registering and querying MCP services and business intelligent agents. The registration center pre-registers agent semantic vectors generated from business capability descriptions of business agents and MCP service semantic vectors generated from description texts of MCP services based on MCP service nodes.

3. The bank card management system based on intelligent agents as described in claim 2, characterized in that, The intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the registration center of the service registration and invocation layer to invoke a business intelligent agent with matching business capabilities and establishes a connection with the invoked business intelligent agent. The general intelligent agent synchronizes the request elements to the business intelligent agent that established the connection; as well as The business intelligence agent that establishes the connection invokes one or more MCP service nodes that match the service capabilities by querying the registration center of the service registration and invocation layer based on the request elements, in order to execute one or more MCP services.

4. The bank card management system based on intelligent agents as described in claim 3, characterized in that, The general-purpose intelligent agent, based on the request intent, invokes a business intelligent agent matching the business capabilities by querying the registry center of the service registration and invocation layer, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the agent registered in the registration center of the service registration and invocation layer.

5. The bank card management system based on intelligent agents as described in claim 4, characterized in that, The business intelligence agent that establishes the connection invokes one or more MCP service nodes that match the service capabilities by querying the service registration and invocation layer's registration center based on the request elements, including: The business intelligence agent that establishes the connection invokes one or more corresponding MCP service nodes based on the similarity between the semantic vector of the request element and the semantic vector of the MCP service registered in the registration center of the service registration and invocation layer.

6. The bank card management system based on intelligent agents as described in claim 5, characterized in that, The intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

7. The bank card management system based on intelligent agents as described in claim 6, characterized in that, The intelligent agent system is configured to further perform the following actions: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

8. The bank card management system based on intelligent agents as described in claim 7, characterized in that, The business intelligence agent includes: A smart card manager agent used to implement bank card management functions; and / or One or more assistive agents used to implement assistive functions.

9. The bank card management system based on intelligent agents as described in claim 8, characterized in that, In the case where the business intelligence agent includes a card management intelligence agent and an auxiliary intelligence agent, The intelligent agent system is configured to perform the following actions: The general-purpose intelligent agent obtains the request intent and request elements by invoking the large language model to parse the natural language request. The general intelligent agent, based on the request intent, queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. The general-purpose intelligent agent synchronizes the request elements to the card manager intelligent agent and the auxiliary intelligent agent that have established the connection; as well as The card manager intelligent agent and the auxiliary intelligent agent that establish the connection query the service registration and invocation layer based on the request elements to invoke multiple MCP service nodes of the business service layer to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of multiple MCP services into natural language execution results using the large language model, and the large language model summarizes them to the card manager intelligent agent that establishes the connection. as well as The card manager intelligent agent that establishes the connection provides the aggregated execution results to the user interaction layer through the general intelligent agent.

10. The bank card management system based on intelligent agents as described in claim 7, characterized in that, The registration center has the following features: The receiving module is configured to receive in advance the registration requests of the one or more MCP service nodes and the one or more business intelligent agents, wherein the registration request includes an MCP service semantic vector generated based on the description text of the MCP service of the MCP service node and an intelligent agent semantic vector generated based on the business capability description of the business intelligent agent. A database is used to store the semantic vectors of the MCP service and the semantic vectors of the agent; The calculation module is used to calculate the first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and to filter out business agents with a first similarity higher than a first threshold as matching results; and to calculate the second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and to filter out MCP service nodes with a second similarity higher than a second threshold as matching results.

11. The bank card management system based on intelligent agents as described in claim 10, characterized in that, The MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. MCP service nodes query the registry center for neighboring MCP service nodes whose semantic vector similarity is higher than the third threshold. as well as MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

12. The bank card management system based on intelligent agents as described in claim 11, characterized in that, The MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

13. A bank card management method based on intelligent agents, implemented using a user interaction layer, a large model service layer, a business service layer, a service registration and invocation layer, and an intelligent agent system, characterized in that... The intelligent agent system includes a general intelligent agent and one or more business intelligent agents. The large model service layer is used to provide a large language model capable of parsing natural language. The business service layer includes one or more MCP service nodes for performing MCP services. The intelligent agent-based bank card management method includes the following steps: The user interaction layer is used to receive natural language requests from users; The general-purpose intelligent agent obtains the natural language request from the user interaction layer and parses the natural language request by calling the large language model to obtain the request intent and request elements. The general-purpose intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent; and The business intelligence agent queries the service registration and invocation layer based on the request elements to invoke one or more MCP service nodes of the business service layer to execute one or more MCP services.

14. The bank card management method based on intelligent agents as described in claim 13, characterized in that, The general-purpose intelligent agent queries the service registration and invocation layer to match and invoke the corresponding business intelligent agent based on the request intent, including: The general agent, based on the request intent, queries the service registration and invocation layer to invoke a business agent with matching business capabilities and establishes a connection with the invoked business agent.

15. The bank card management method based on intelligent agents as described in claim 14, characterized in that, Further includes: The business intelligence agent that establishes the connection converts the execution result of one MCP service, or the combined execution result of multiple MCP services, into a natural language execution result using the large language model.

16. The bank card management method based on intelligent agents as described in claim 15, characterized in that, Further includes: The business intelligence agent that establishes the connection will further provide the execution result of the natural language to the user interaction layer through the general intelligence agent.

17. The bank card management method based on intelligent agents as described in claim 16, characterized in that, The general-purpose intelligent agent, based on the request intent, queries the service registration and invocation layer to invoke a business intelligent agent that matches the business capabilities, including: The general-purpose agent invokes the corresponding business agent based on the similarity between the semantic vector of the request intent and the semantic vector of the business capabilities of the business agents already registered in the service registration and invocation layer.

18. The bank card management method based on intelligent agents as described in claim 17, characterized in that, The business intelligent agents include: a card manager intelligent agent for implementing bank card management functions; and one or more auxiliary intelligent agents for implementing auxiliary functions. Based on the request intent, the general intelligent agent queries the service registration and invocation layer to invoke the card manager intelligent agent and auxiliary intelligent agent that match the business capabilities, and establishes a connection with the invoked card manager intelligent agent and auxiliary intelligent agent. Furthermore, the general intelligent agent synchronizes the request elements to the established card manager intelligent agent and auxiliary intelligent agent. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection call each invoke multiple MCP service nodes of the business service layer by querying the service registration and invocation layer based on the request elements to execute multiple MCP services. The card manager intelligent agent and the auxiliary intelligent agent that establish the connection convert the execution results of the multiple MCP services into natural language execution results using the large language model, and the large language model summarizes them to the card manager intelligent agent that establishes the connection. The card manager intelligent agent that establishes the connection provides the summarized execution results to the user interaction layer through the general intelligent agent.

19. The bank card management method based on intelligent agents as described in claim 18, characterized in that, The service registration and invocation layer has a registration center for registering and querying MCP services and business intelligent agents.

20. The registry center is configured to perform the following actions: The system receives registration requests in advance from one or more MCP service nodes and one or more business intelligent agents, wherein... The registration request includes an MCP service semantic vector generated from the description text of the MCP service based on the MCP service node, and an agent semantic vector generated from the description of the business capabilities of the business agent. Store the MCP service semantic vector and the agent semantic vector; The system calculates a first similarity between the semantic vector of the request intent and the semantic vector of the agent stored in the database, and selects business agents with a first similarity higher than a first threshold as matching results. It also calculates a second similarity between the semantic vector of the request intent and the semantic vector of the MCP service stored in the database, and selects MCP service nodes with a second similarity higher than a second threshold as matching results.

21. The bank card management method based on intelligent agents as described in claim 19, characterized in that, The MCP service node in the business service layer is configured to perform the following actions: MCP service nodes use semantic embedding models to generate semantic vectors for their own MCP service description text. The semantic vector of the MCP service and the corresponding node metadata are sent to the registration center for registration. MCP service nodes query the registry center for neighboring MCP service nodes whose semantic vector similarity is higher than the third threshold. as well as MCP service nodes establish connections with their neighboring MCP service nodes and aggregate with them to form a cluster.

22. The bank card management method based on intelligent agents as described in claim 20, characterized in that, The MCP service node in the business service layer is configured to further perform the following actions: Each MCP service node maintains semantic vector similarity with neighboring MCP service nodes. When the semantic vector similarity is lower than the third threshold, the aggregated cluster is disbanded and re-aggregation is triggered.

23. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the agent-based bank card management method according to any one of claims 13 to 21.

24. A computer device, comprising a storage module, a processor, and a computer program stored on the storage module and executable on the processor, characterized in that, When the processor executes the computer program, it implements the agent-based bank card management method according to any one of claims 13 to 21.

25. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the agent-based bank card management method according to any one of claims 13 to 21.

Images