Intelligent agent and tool chain collaborative data processing method, equipment and medium

Abstract

The invention discloses an agent and tool chain collaborative data processing method and device and a medium, and relates to the technical field of agent application. The method comprises the steps of receiving a data processing task target and related configuration parameters input by a user, and initializing task context data based on the task target and the configuration parameters; executing a circular processing process by an agent based on a ReAct framework, and dynamically adjusting a task strategy by iteratively executing a thinking stage, an action stage and an observation stage until a task completion condition is met; in the circular processing process, based on standardized tool calling instruction data generated by the intelligent agent, cooperatively calling the MCP tool chain and a plurality of tools in the external tool adaptation layer to execute a data processing task; performing aggregation processing and data verification on the multi-source result data returned by the tool execution to generate integrated analysis result data; and displaying task execution process data and a final analysis result in real time through a front-end interaction interface, and dynamically adjusting a task execution process based on user intervention instruction data.

Description

Technical Field

[0001] This application relates to the field of intelligent agent application technology, and in particular to a data processing method, device and medium for intelligent agents and toolchain collaboration. Background Technology

[0002] In the current field of enterprise data processing and analysis, pre-defined linear processes or scripts are commonly used to invoke toolchains to execute tasks. These systems typically rely on pre-programmed, fixed execution paths, lacking the ability to dynamically assess and make autonomous decisions based on intermediate results generated during task execution. When dealing with complex objectives or diverse data sources, this rigid process struggles to adapt to changing needs, resulting in a severe lack of flexibility and intelligence in task execution.

[0003] Furthermore, existing systems often face challenges in coordinating internal and external tools. Due to differing interface specifications and data formats among various tools, seamless command transmission and result interaction are difficult to achieve. This "silo" phenomenon between tools easily leads to gaps in data processing workflows, reducing task execution efficiency and limiting the possibility of comprehensive utilization of multi-source data for in-depth analysis.

[0004] Therefore, how to build a data processing system that can dynamically adjust strategies and seamlessly coordinate multiple tools has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This application provides a data processing method, device, and medium for collaboration between intelligent agents and toolchains to solve the following technical problem: how to build a data processing system that can dynamically adjust strategies and seamlessly coordinate multiple tools.

[0006] In a first aspect, embodiments of this application provide a data processing method for collaboration between an intelligent agent and a toolchain. The method includes: receiving a data processing task objective and related configuration parameters input by a user, and initializing task context data based on the task objective and the configuration parameters; an intelligent agent based on the ReAct framework executing a cyclic processing process, dynamically adjusting the task strategy through iterative execution of a thinking phase, an action phase, and an observation phase until the task completion conditions are met; during the cyclic processing, based on standardized tool call instruction data generated by the intelligent agent, collaboratively calling multiple tools in the MCP toolchain and external tool adaptation layer to execute data processing tasks; aggregating and verifying the multi-source result data returned by the tool execution to generate integrated analysis result data; and displaying the task execution process data and the final analysis results in real time through a front-end interactive interface, and dynamically adjusting the task execution process based on user intervention instruction data.

[0007] In one embodiment of this application, an agent based on the ReAct framework executes a cyclical processing procedure, dynamically adjusting the task strategy through iterative execution of a thinking phase, an action phase, and an observation phase until the task completion conditions are met. Specifically, this includes: in the thinking phase, generating structured thinking logic data based on task target data, historical action record data, and tool return result data in the current task context data, and determining the next action data based on a decision tree algorithm; in the action phase, converting the next action data into standardized tool call instruction data; in the observation phase, parsing, formatting, and scoring the tool result data returned after tool execution to generate structured observation data, and updating the historical record data and tool return result data in the task context data; iteratively executing the thinking phase, the action phase, and the observation phase until the agent determines that the task is completed based on the task completion conditions; wherein, the task completion conditions include an analysis result completeness index and a user-defined termination condition.

[0008] In one embodiment of this application, structured thinking logic data is generated based on task target data, historical action record data, and tool return result data in the current task context data. Specifically, this includes: parsing the task target data using natural language processing technology, extracting key task elements, and generating task intent data; wherein the task intent data includes data acquisition requirements and analysis and verification requirements; analyzing historical action record data using sequence modeling technology, identifying tool call patterns and parameter dependencies, and generating action sequence pattern data; evaluating tool return result data using a logical reasoning engine, calculating data integrity and consistency indicators, and generating current data status evaluation data; integrating the task intent data, action sequence pattern data, and current data status evaluation data, and generating the structured thinking logic data using a decision tree algorithm; wherein the structured thinking logic data includes a data acquisition step sequence, an analysis and verification step sequence, and a conclusion generation step.

[0009] In one embodiment of this application, converting the next action data into standardized tool invocation instruction data specifically includes: parsing action type data and action parameter data in the next action data; wherein, the action type data includes data query action, analysis action, or visualization action; mapping the action type data to predefined tool type data; wherein the tool type data corresponds to data retrieval tools, analysis tools, visualization tools, or reporting tools in the MCP toolchain, and web search tools in the external tool adaptation layer; generating standardized tool invocation instruction data based on the action parameter data and the tool type data; wherein, the standardized tool invocation instruction data includes tool identifier data, invocation parameter list data, and request identifier data; verifying the completeness and validity of the standardized tool invocation instruction data, checking the completeness and format compliance of required parameters, and outputting the verified standardized tool invocation instruction data for tool invocation.

[0010] In one embodiment of this application, the tool result data returned after tool execution is parsed, formatted, and scored with confidence to generate structured observation data. Specifically, this includes: receiving raw result data returned by the tool execution; wherein the raw result data includes text data, numerical data, or image data; parsing the raw result data to extract key information data and formatting the raw result data into intermediate structure data; scoring the key information data with confidence based on multi-source validation rules to generate confidence-labeled data; wherein the multi-source validation rules include cross-validation of data sources and logical consistency checks; and combining the intermediate structure data and the confidence-labeled data to generate structured observation data.

[0011] In one embodiment of this application, based on standardized tool call instruction data generated by the intelligent agent, multiple tools in the MCP toolchain and external tool adaptation layer are collaboratively invoked to perform data processing tasks. Specifically, this includes: parsing the tool identifier data and call parameter data in the standardized tool call instruction data, and routing the call request to the corresponding MCP tool service or external tool adaptation service based on the service registry; for the MCP toolchain, executing data query tasks, data analysis tasks, data visualization tasks, or report generation tasks, and generating first tool result data; for the external tool adaptation layer, executing web search tasks, data crawling tasks, or API call tasks, and performing multi-source verification and logical consistency checks on the returned raw data to generate second tool result data; integrating the first tool result data and the second tool result data to generate a multi-source tool result dataset, and feeding it back to the intelligent agent for updating task context data.

[0012] In one embodiment of this application, the multi-source result data returned by the tool execution is aggregated and validated to generate integrated analysis result data. Specifically, this includes: aggregating multi-source tool result data from the MCP toolchain and external tool adaptation layers, performing data alignment based on timestamp data and task identifier data to generate an initial aggregated dataset; performing data cleaning on the initial aggregated dataset to generate a cleaned dataset; performing consistency validation on the cleaned dataset, using a rule engine to check the logical consistency and range validity of the data, marking low-confidence data entries, and generating a data quality assessment report; and based on the data quality assessment report, performing structure transformation and format unification on the validated data to generate integrated analysis result data.

[0013] In one embodiment of this application, task execution process data and final analysis results are displayed in real time through a front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention command data. Specifically, this includes: receiving task execution process data through a real-time communication interface; wherein the task execution process data includes structured thinking logic data, tool call progress data, and intermediate result data; rendering the task execution process data using visualization components in the front-end interface; wherein the thinking flow component converts the structured thinking logic data into a readable text stream, the progress bar component dynamically displays the tool call progress, and the chart component visualizes the intermediate result data; receiving intervention command data input by the user through interactive components; wherein the interactive components include parameter input boxes and action selection buttons, and the intervention command data includes parameter adjustment commands, action addition commands, or task redirection commands; parsing the intervention command data, generating task adjustment data, and sending the task adjustment data to the intelligent agent to update the task target data, reinitializing the task context data, and triggering a new round of cyclic processing.

[0014] Secondly, embodiments of this application also provide a data processing device for collaboration between an intelligent agent and a toolchain. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a data processing method for collaboration between an intelligent agent and a toolchain as described above.

[0015] Thirdly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement a data processing method for collaboration between an intelligent agent and a toolchain as described above.

[0016] The data processing method, device, and medium for intelligent agent and toolchain collaboration provided in this application have the following beneficial effects: Through an intelligent agent cyclic processing mechanism based on the ReAct framework, dynamic generation and real-time adjustment of task strategies are achieved, significantly improving the intelligence and execution flexibility of complex data processing tasks; By using standardized instructions to collaboratively call the MCP toolchain and external tools, data and instruction barriers in multi-tool collaboration are effectively overcome, enhancing the system's integration and scalability; Simultaneously, through the aggregation and verification of multi-source result data and real-time visualization interaction of the front-end interface, not only is the accuracy and reliability of the final analysis results ensured, but the entire task process is also made visible, traceable, and controllable, enabling users to deeply participate in the analysis process, thereby greatly improving the overall efficiency, quality, and user experience of data analysis. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a data processing method for collaboration between an intelligent agent and a toolchain, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of a data processing device that enables collaboration between an intelligent agent and a toolchain, as provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. 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.

[0019] This application provides a data processing method, device, and medium for collaboration between intelligent agents and toolchains, in order to solve the technical problem of how to build a data processing system that can dynamically adjust strategies and seamlessly coordinate multiple tools.

[0020] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating a data processing method for collaboration between an intelligent agent and a toolchain, provided as an embodiment of this application. Figure 1 As shown in the figure, the data processing method for collaboration between an intelligent agent and a toolchain provided in this application embodiment specifically includes the following steps: Step 101: Receive the data processing task objective and related configuration parameters input by the user, and initialize the task context data based on the task objective and the configuration parameters.

[0022] In this embodiment, the "data processing task objective" refers to the specific analysis requirements or problem description input by the user through the front-end interactive interface (e.g., a web application developed based on the React framework), usually expressed in natural language, such as "analyze the reasons for the decline in sales of financial product A this quarter" input by the user in a financial data analysis scenario. This objective is captured by the front-end input component (e.g., a text box or voice recognition interface) and transmitted to the back-end processing system via HTTP or WebSocket protocol. The "related configuration parameters" are auxiliary constraints or preference options set to achieve this objective, including data time range, visualization type, data source priority, etc. These parameters are customized or input by the user through form elements in the front-end interface (e.g., drop-down menus, checkboxes, or sliders). After the system receives the original input, the back-end service calls the parsing module to standardize the task objective and configuration parameters, for example, using natural language processing technology to extract key task elements (e.g., "product A", "sales", "reasons for decline"), and converting the configuration parameters into a structured data format to ensure the accurate execution of subsequent steps.

[0023] Understandably, this parsing process helps eliminate ambiguity in user input and lays the foundation for initializing task context data. Specifically, "initializing task context data" refers to the system creating a dynamically updatable data object that persists throughout the entire task lifecycle and serves as the core basis for agent decision-making. The initialization process includes: generating a unique task identifier (such as a UUID) to distinguish different task instances; storing the parsed task objective description and structured configuration parameters; initializing an empty historical action record list to sequentially record actions executed in subsequent loops and their metadata; and initializing an empty tool return result list to accumulate intermediate data generated after each tool execution. It should be noted that task context data is typically persisted in an in-memory database (such as Redis) or a relational database to ensure consistency of task state during system restarts or fault recovery. For example, in a financial data analysis scenario, the initialized task context might include a task ID "TASK_001", an objective description "analyze the reasons for the decline in sales of financial product A", configuration parameters such as "time range: 2023Q1-Q2", "visualization preference: trend chart", and an empty historical record and result list.

[0024] Step 102: The agent based on the ReAct framework executes a cyclical processing procedure, dynamically adjusting the task strategy by iteratively executing the thinking phase, action phase, and observation phase until the task completion conditions are met.

[0025] In this embodiment, the ReAct framework constitutes the core reasoning mechanism of the agent, which realizes the dynamic evolution of the task strategy through the iterative cycle of "thinking-action-observation".

[0026] In one embodiment of this application, an agent based on the ReAct framework executes a cyclical processing procedure, dynamically adjusting the task strategy through iterative execution of a thinking phase, an action phase, and an observation phase until the task completion conditions are met. Specifically, this includes: in the thinking phase, generating structured thinking logic data based on task target data, historical action record data, and tool return result data in the current task context data, and determining the next action data based on a decision tree algorithm; in the action phase, converting the next action data into standardized tool call instruction data; in the observation phase, parsing, formatting, and scoring the tool result data returned after tool execution to generate structured observation data, and updating the historical record data and tool return result data in the task context data; iteratively executing the thinking phase, the action phase, and the observation phase until the agent determines that the task is completed based on the task completion conditions; wherein, the task completion conditions include an analysis result completeness index and a user-defined termination condition.

[0027] In one embodiment of this application, structured thinking logic data is generated based on task target data, historical action record data, and tool return result data in the current task context data. Specifically, this includes: parsing the task target data using natural language processing technology, extracting key task elements, and generating task intent data; wherein the task intent data includes data acquisition requirements and analysis and verification requirements; analyzing historical action record data using sequence modeling technology, identifying tool call patterns and parameter dependencies, and generating action sequence pattern data; evaluating tool return result data using a logical reasoning engine, calculating data integrity and consistency indicators, and generating current data status evaluation data; integrating the task intent data, action sequence pattern data, and current data status evaluation data, and generating the structured thinking logic data using a decision tree algorithm; wherein the structured thinking logic data includes a data acquisition step sequence, an analysis and verification step sequence, and a conclusion generation step.

[0028] In one embodiment of this application, converting the next action data into standardized tool invocation instruction data specifically includes: parsing action type data and action parameter data in the next action data; wherein, the action type data includes data query action, analysis action, or visualization action; mapping the action type data to predefined tool type data; wherein the tool type data corresponds to data retrieval tools, analysis tools, visualization tools, or reporting tools in the MCP toolchain, and web search tools in the external tool adaptation layer; generating standardized tool invocation instruction data based on the action parameter data and the tool type data; wherein, the standardized tool invocation instruction data includes tool identifier data, invocation parameter list data, and request identifier data; verifying the completeness and validity of the standardized tool invocation instruction data, checking the completeness and format compliance of required parameters, and outputting the verified standardized tool invocation instruction data for tool invocation.

[0029] In one embodiment of this application, the tool result data returned after tool execution is parsed, formatted, and scored with confidence to generate structured observation data. Specifically, this includes: receiving raw result data returned by the tool execution; wherein the raw result data includes text data, numerical data, or image data; parsing the raw result data to extract key information data and formatting the raw result data into intermediate structure data; scoring the key information data with confidence based on multi-source validation rules to generate confidence-labeled data; wherein the multi-source validation rules include cross-validation of data sources and logical consistency checks; and combining the intermediate structure data and the confidence-labeled data to generate structured observation data.

[0030] Understandably, during the thinking phase, the agent performs comprehensive reasoning based on task objective data, historical action record data, and tool return result data within the current task context data. For example, this phase uses natural language processing technology to parse the task objective to extract key task elements, generating task intent data that includes data acquisition and analysis verification needs. Simultaneously, it analyzes historical action records based on sequence modeling technology to identify tool call patterns, and combines a logical reasoning engine to evaluate the data integrity and consistency of the tool's returned results. Finally, it integrates the above information through a decision tree algorithm to generate structured thinking logic data, which explicitly includes the next action decision, such as "We need to first obtain product sales data and then conduct trend analysis."

[0031] The system then enters the action phase, where it transforms the next action data determined in the thinking phase into standardized tool invocation instruction data. Specifically, this phase first parses the action type data (such as data query or analysis actions) and action parameter data, then maps them to predefined tool type data (such as data retrieval tools or analysis tools in the MCP toolchain). Finally, it generates standardized instructions containing tool identifier data, invocation parameter list data, and request identifier data, and verifies their completeness and format compliance before sending. It should be noted that this standardization conversion process is the key technological foundation for enabling collaborative invocation of multiple tools.

[0032] Following the observation phase, the system receives the raw result data (including heterogeneous data such as text, numerical values, or images) returned after the tool's execution. It parses this data to extract key information and formats it into intermediate structured data. Simultaneously, it generates confidence-labeled data based on multi-source validation rules, and finally combines this data to generate structured observation data, updating the historical records and tool return result list in the task context data. For example, when an external search tool returns information about market policies from multiple sources, the observation phase assesses the credibility of the information through cross-validation and logical consistency checks, eliminating contradictory low-confidence data. These three phases are executed cyclically until the agent determines task completion based on the completeness of the analysis results (e.g., all key analytical dimensions are covered and the conclusions are consistent) or a user-defined termination condition. This cyclical mechanism allows the agent to dynamically adjust its strategy based on real-time feedback; for instance, if the initial analysis fails to find obvious anomalies, it automatically triggers deeper data exploration, thus overcoming the limitations of traditional fixed processes.

[0033] Step 103: During the loop processing, based on the standardized tool call instruction data generated by the agent, multiple tools in the MCP toolchain and external tool adaptation layer are collaboratively invoked to perform data processing tasks.

[0034] In this embodiment, this step is a key step in realizing the collaborative execution of multiple tools. Its core lies in transforming the agent's decisions into specific tool execution actions through a unified instruction parsing and routing mechanism.

[0035] In one embodiment of this application, based on standardized tool call instruction data generated by the intelligent agent, multiple tools in the MCP toolchain and external tool adaptation layer are collaboratively invoked to perform data processing tasks. Specifically, this includes: parsing the tool identifier data and call parameter data in the standardized tool call instruction data, and routing the call request to the corresponding MCP tool service or external tool adaptation service based on the service registry; for the MCP toolchain, executing data query tasks, data analysis tasks, data visualization tasks, or report generation tasks, and generating first tool result data; for the external tool adaptation layer, executing web search tasks, data crawling tasks, or API call tasks, and performing multi-source verification and logical consistency checks on the returned raw data to generate second tool result data; integrating the first tool result data and the second tool result data to generate a multi-source tool result dataset, and feeding it back to the intelligent agent for updating task context data.

[0036] In this embodiment, the system first parses the tool identifier data and call parameter data in the standardized tool call instruction data. Then, based on a pre-configured service registry, it routes the call request to the corresponding MCP tool service or external tool adaptation service. For example, when the tool identifier in the instruction points to "MCP-ANALYSIS-001", the system routes it to the internal analysis tool service; if the identifier is "EXT-SEARCH-001", it routes it to the external search adaptation service. It should be noted that this service registry-based routing mechanism ensures the accuracy of tool calls and the scalability of the system. When a new tool is added, it only needs to be registered in the registry to be discovered and invoked by the agent.

[0037] During the execution of the MCP toolchain, the system performs corresponding data processing tasks based on the tool type and parameters: the data retrieval tool executes SQL queries or API calls through a pre-configured database connection interface to obtain raw data such as product sales data and user profile data from the business system; the analysis tool loads models such as time series analysis and correlation analysis to calculate the input data and output analysis conclusions and key indicators; the visualization tool generates chart data objects based on the analysis results; and the reporting tool integrates all intermediate results to generate a structured report containing data, analysis process, and recommended measures. Understandably, these tools receive instructions and return structured results through standardized interfaces, ensuring the smooth flow of data within the toolchain.

[0038] For the execution of the external tool adaptation layer, the system calls web search tools, data scraping tools, or third-party API interfaces to obtain external data, and performs crucial data verification processing before returning the results. For example, when obtaining macroeconomic policy or competitor dynamic data, the external tool adaptation layer will query at least three authoritative sources in parallel, assess the credibility of the data through cross-validation and logical consistency checks, and finally generate result data with confidence scores. It should be noted that this multi-source verification mechanism significantly improves the reliability of external data and avoids the impact of erroneous data on the analysis conclusions.

[0039] Finally, the system integrates the first tool result data returned by the MCP toolchain with the second tool result data returned by the external tool adaptation layer to form a complete multi-source tool result dataset, and immediately feeds it back to the agent to update the task context data.

[0040] Step 104: Aggregate and validate the multi-source results data returned by the tool to generate integrated analysis results data.

[0041] In this embodiment, this step is crucial in transforming the results of distributed and heterogeneous tool executions into a unified and reliable analytical dataset.

[0042] In one embodiment of this application, the multi-source result data returned by the tool execution is aggregated and validated to generate integrated analysis result data. Specifically, this includes: aggregating multi-source tool result data from the MCP toolchain and external tool adaptation layers, performing data alignment based on timestamp data and task identifier data to generate an initial aggregated dataset; performing data cleaning on the initial aggregated dataset to generate a cleaned dataset; performing consistency validation on the cleaned dataset, using a rule engine to check the logical consistency and range validity of the data, marking low-confidence data entries, and generating a data quality assessment report; and based on the data quality assessment report, performing structure transformation and format unification on the validated data to generate integrated analysis result data.

[0043] In this embodiment, the system first aggregates multi-source tool result data from the MCP toolchain and the external tool adaptation layer. Based on the timestamp and task identifier data carried in each result data, it performs spatiotemporal alignment, linking and integrating related data entries belonging to the same task period to form an initial aggregated dataset. For example, in a financial product analysis scenario, the system will correlate and match sales data returned by the data retrieval tool, trend analysis results output by the analysis tool, and market policy information obtained by the external search tool according to the time dimension and task ID, ensuring that all data is within a unified analysis framework. It should be noted that this metadata-based aggregation mechanism effectively solves the problem of inconsistent data timing caused by asynchronous execution of multiple tools.

[0044] Subsequently, the system performs data cleaning on the initial aggregated dataset. This process includes handling missing values, removing obvious outliers, extracting key information from textual results, and standardizing their representation. It's understandable that data cleaning not only focuses on the surface quality of the data but also incorporates in-depth processing based on business logic. For example, when extreme values ​​in sales data exceed historical fluctuation ranges, the system marks them as data to be verified and focuses on checking them in subsequent verification stages. After cleaning, the system uses a rule engine to perform consistency checks on the cleaned dataset, checking the logical relationships between data and the reasonableness of numerical values. Specifically, the rule engine verifies the consistency of data from different sources, such as whether the downward trend reflected in internal sales data matches the conclusions of external market analysis. It also checks the validity of the range of numerical data, such as whether the rate of return is within a reasonable range. It should be noted that low-confidence data entries discovered during this process are specially marked and recorded in the data quality assessment report.

[0045] Finally, guided by the data quality assessment report, the system performs structured transformation and format standardization on the validated data, generating the final integrated analysis results. For example, in a financial analysis scenario, this process converts data in different formats, such as numerical sales data, text-based analysis conclusions, and chart-based visualizations, into a standardized JSON-formatted structured data object, containing complete data fields, quality labels, and metadata information. Understandably, this rigorous aggregation and validation process not only ensures the reliability of the final analysis results but also provides high-quality, standardized data input for subsequent result presentation and report generation, thereby guaranteeing the credibility and decision-making reference value of the entire system's output conclusions.

[0046] Step 105: Display the task execution process data and final analysis results in real time through the front-end interactive interface, and dynamically adjust the task execution process based on user intervention command data.

[0047] In this embodiment, this step establishes a key channel for two-way interaction between the user and the intelligent system, enabling visualization and intervention of the processing procedure.

[0048] In one embodiment of this application, task execution process data and final analysis results are displayed in real time through a front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention command data. Specifically, this includes: receiving task execution process data through a real-time communication interface; wherein the task execution process data includes structured thinking logic data, tool call progress data, and intermediate result data; rendering the task execution process data using visualization components in the front-end interface; wherein the thinking flow component converts the structured thinking logic data into a readable text stream, the progress bar component dynamically displays the tool call progress, and the chart component visualizes the intermediate result data; receiving intervention command data input by the user through interactive components; wherein the interactive components include parameter input boxes and action selection buttons, and the intervention command data includes parameter adjustment commands, action addition commands, or task redirection commands; parsing the intervention command data, generating task adjustment data, and sending the task adjustment data to the intelligent agent to update the task target data, reinitializing the task context data, and triggering a new round of cyclic processing.

[0049] In this embodiment, the system continuously acquires dynamic data during task execution through a real-time communication interface, including structured thinking logic data generated during the thinking phase, tool call progress data generated during the action phase, and various intermediate result data returned during the observation phase. For example, when the agent is analyzing sales data of financial products, the front-end interface uses a thinking flow component to convert the agent's reasoning process into a readable text stream, such as "Acquiring sales data for the past three months → Confirmed downward sales trend → Starting to analyze user profile correlation." Simultaneously, a progress bar component dynamically displays the execution status of each tool call, while a chart component renders generated sales trend charts, user distribution charts, and other visualizations in real time. It should be noted that this comprehensive process display makes the traditional "black box" data processing completely transparent, allowing users to clearly understand the system's decision-making logic and execution path.

[0050] Understandably, the front-end's interactive capabilities extend beyond passive display; more importantly, it provides a channel for proactive intervention. The system receives user intervention command data through specially designed interactive components. These components include parameter input boxes for adjusting analysis parameters, action selection buttons for adding analysis dimensions, and drop-down menus for redirecting task flows. For example, when a user observes preliminary analysis results, they can click the "Supplement Regional Dimension Analysis" button or select "Add Competitive Analysis" from the drop-down menu. The system immediately converts these operations into structured intervention command data. Subsequently, the system parses this intervention command data, accurately identifies the user's intent type and specific parameters, generates corresponding task adjustment data, and transmits this adjustment data to the agent layer in real time. It should be noted that this intervention mechanism triggers a dynamic restructuring of the task execution flow: upon receiving the adjustment command, the agent immediately updates the task target data, reinitializes the task context data, and initiates a new round of cyclical processing based on the new task requirements. This design enables the system to flexibly respond to users' real-time needs, forming an enhanced intelligent closed loop of human-machine collaboration. It ensures the automated efficiency of system processing while retaining the decision-making leadership of human experts. Finally, the system presents the refined analysis conclusions, visualization charts, and complete reports to users in an interactive manner through the result rendering component, completing the entire data processing workflow.

[0051] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a data processing device for collaboration between an intelligent agent and a toolchain, the structure of which is as follows: Figure 2 As shown.

[0052] Figure 2 This is a schematic diagram of the internal structure of a data processing device that enables collaboration between an intelligent agent and a toolchain, as provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions executable by at least one processor, which are executed by at least one processor 201 to enable at least one processor 201 to: Receive user input of data processing task objectives and related configuration parameters, and initialize task context data based on the task objectives and configuration parameters; Based on the ReAct framework, the agent executes a cyclical processing process, dynamically adjusting the task strategy by iteratively executing the thinking phase, action phase, and observation phase until the task completion conditions are met. During the cyclic processing, based on the standardized tool call instruction data generated by the intelligent agent, multiple tools in the MCP toolchain and the external tool adaptation layer are collaboratively invoked to perform data processing tasks. The multi-source results data returned by the tool are aggregated and validated to generate integrated analysis results data. The task execution process data and final analysis results are displayed in real time through the front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention command data.

[0053] Some embodiments of this application provide corresponding to Figure 1 A computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: Receive user input of data processing task objectives and related configuration parameters, and initialize task context data based on the task objectives and configuration parameters; Based on the ReAct framework, the agent executes a cyclical processing process, dynamically adjusting the task strategy by iteratively executing the thinking phase, action phase, and observation phase until the task completion conditions are met. During the cyclic processing, based on the standardized tool call instruction data generated by the intelligent agent, multiple tools in the MCP toolchain and the external tool adaptation layer are collaboratively invoked to perform data processing tasks. The multi-source results data returned by the tool are aggregated and validated to generate integrated analysis results data. The task execution process data and final analysis results are displayed in real time through the front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention command data.

[0054] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0055] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0056] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0061] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0062] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0063] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0064] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A data processing method for collaboration between an intelligent agent and a toolchain, characterized in that, The method includes: Receive user input of data processing task objectives and related configuration parameters, and initialize task context data based on the task objectives and configuration parameters; Based on the ReAct framework, the agent executes a cyclical processing process, dynamically adjusting the task strategy by iteratively executing the thinking phase, action phase, and observation phase until the task completion conditions are met. During the cyclic processing, based on the standardized tool call instruction data generated by the agent, multiple tools in the MCP toolchain and the external tool adaptation layer are collaboratively invoked to perform data processing tasks. The multi-source results data returned by the tool are aggregated and validated to generate integrated analysis results data. The task execution process data and final analysis results are displayed in real time through the front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention command data.

2. The data processing method for collaboration between intelligent agents and toolchains according to claim 1, characterized in that, Based on the ReAct framework, the agent executes a cyclical processing procedure, dynamically adjusting the task strategy through iterative execution of the thinking phase, action phase, and observation phase until the task completion conditions are met. Specifically, this includes: During the thinking phase, structured thinking logic data is generated based on the task objective data, historical action record data, and tool return result data in the current task context data, and the next action data is determined based on the decision tree algorithm; During the action phase, the data for the next action is converted into standardized tool invocation instruction data; During the observation phase, the tool result data returned after the tool is executed is parsed, formatted, and scored with confidence to generate structured observation data, and the historical data and tool return result data in the task context data are updated. The thinking phase, the action phase, and the observation phase are executed iteratively until the agent determines that the task is completed based on the task completion conditions; wherein, the task completion conditions include the completeness index of the analysis results and user-defined termination conditions.

3. The data processing method for collaboration between intelligent agents and toolchains according to claim 2, characterized in that, Based on the task objective data, historical action record data, and tool return result data in the current task context data, structured thinking logic data is generated, specifically including: The task objective data is analyzed using natural language processing technology to extract key task elements and generate task intent data; wherein, the task intent data includes data acquisition requirements and analysis and verification requirements; Based on sequence modeling technology, historical action record data is analyzed to identify tool call patterns and parameter dependencies, and action sequence pattern data is generated. Based on the results data returned by the logical reasoning engine evaluation tool, calculate the data integrity and consistency indicators, and generate the current data status evaluation data; By integrating task intent data, action sequence pattern data, and current data status assessment data, the structured thinking logic data is generated using a decision tree algorithm; wherein, the structured thinking logic data includes a data acquisition step sequence, an analysis and verification step sequence, and a conclusion generation step.

4. The data processing method for collaboration between intelligent agents and toolchains according to claim 2, characterized in that, The next action data is converted into standardized tool call command data, specifically including: The action type data and action parameter data in the next action data are analyzed; wherein, the action type data includes data query action, analysis action, or visualization action; Map the action type data to predefined tool type data; wherein the tool type data corresponds to data retrieval tools, analysis tools, visualization tools or reporting tools in the MCP toolchain, as well as web search tools in the external tool adaptation layer; Based on the action parameter data and the tool type data, standardized tool invocation instruction data is generated; wherein, the standardized tool invocation instruction data includes tool identifier data, invocation parameter list data, and request identifier data; Verify the completeness and validity of the standardized tool call command data, check the completeness and format compliance of the required parameters, and output the verified standardized tool call command data for tool invocation.

5. The data processing method for collaboration between intelligent agents and toolchains according to claim 2, characterized in that, The tool results data returned after execution are parsed, formatted, and scored with confidence to generate structured observation data, specifically including: The receiving tool returns raw result data; wherein, the raw result data includes text data, numerical data, or image data; The original result data is parsed to extract key information data, and the original result data is formatted into intermediate structure data. The key information data is scored with confidence based on multi-source verification rules to generate confidence-labeled data; wherein, the multi-source verification rules include cross-validation of data sources and logical consistency checks; The intermediate structure data and the confidence level labeled data are combined to generate structured observation data.

6. The data processing method for collaboration between intelligent agents and toolchains according to claim 1, characterized in that, Based on standardized tool call command data generated by the intelligent agent, multiple tools in the MCP toolchain and external tool adaptation layer are collaboratively invoked to perform data processing tasks, specifically including: The tool identifier data and call parameter data in the standardized tool call instruction data are parsed, and the call request is routed to the corresponding MCP tool service or external tool adaptation service based on the service registry. For the MCP toolchain, perform data query tasks, data analysis tasks, data visualization tasks, or report generation tasks, and generate the first tool result data; For the external tool adaptation layer, perform web search tasks, data crawling tasks, or API call tasks, and perform multi-source verification and logical consistency checks on the returned raw data to generate second tool result data; The first tool result data and the second tool result data are integrated to generate a multi-source tool result dataset, which is then fed back to the agent to update the task context data.

7. The data processing method for collaboration between intelligent agents and toolchains according to claim 1, characterized in that, The multi-source results returned by the tool are aggregated and validated to generate integrated analysis results, which include: Aggregate multi-source tool result data from the MCP toolchain and external tool adaptation layer, and perform data alignment based on timestamp data and task identifier data to generate an initial aggregated dataset; Perform data cleaning operations on the initial aggregated dataset to generate a cleaned dataset; The cleaned dataset is subjected to consistency verification. A rule engine is used to check the logical consistency and range validity of the data, low-confidence data entries are marked, and a data quality assessment report is generated. Based on the data quality assessment report, the verified data undergoes structure transformation and format standardization to generate integrated analysis results.

8. The data processing method for collaboration between intelligent agents and toolchains according to claim 1, characterized in that, The task execution process data and final analysis results are displayed in real time through a front-end interactive interface, and the task execution process is dynamically adjusted based on user intervention commands. Specifically, this includes: The task execution process data is received through a real-time communication interface; wherein, the task execution process data includes structured thinking logic data, tool call progress data, and intermediate result data; The task execution process data is rendered using visualization components in the front-end interface; among them, the thinking flow component converts structured thinking logic data into a readable text stream, the progress bar component dynamically displays the tool call progress, and the chart component visualizes intermediate result data; Receive intervention instruction data input by the user through an interactive component; wherein the interactive component includes a parameter input box and an action selection button, and the intervention instruction data includes parameter adjustment instructions, action addition instructions, or task redirection instructions; The intervention instruction data is parsed to generate task adjustment data, which is then sent to the agent to update the task target data, reinitialize the task context data, and trigger a new round of loop processing.

9. A data processing device for collaboration between intelligent agents and toolchains, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a data processing method for collaboration between an agent and a toolchain as described in any one of claims 1-8.

10. A computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, a data processing method for collaboration between an intelligent agent and a toolchain as described in any one of claims 1-8 is implemented.

Images