The Intelligent Analysis Agent provides you with MCP service integration and management capabilities. Through this feature, you can connect and manage external MCP services to expand DataAgent's data acquisition scope (typically external data, such as online search, third-party platform data, and more) and analytical capabilities (to enhance the agent's data analysis abilities, such as attribution, prediction, and more). This document introduces the MCP service management features and detailed operation instructions.

The Intelligent Analysis Agent provides both system-level and agent-level MCP integration capabilities. A comparison of these two types of MCP services is provided below.

After integrating MCP services, when using features such as in-depth research and insight reports, the large model will determine, based on your questions and the MCP service configuration, whether it needs to use MCP services to further improve the quality of analysis results.

• The Intelligent Analysis Agent supports MCP service functionality starting from version 3.7.1.
• The current Smart Q&A feature does not support the use of integrated MCP services. MCP services can be used in the in-depth research and insight report features.
• The MCP protocol version of any custom MCP service added must be 2025-06-18 or later.

Except for the applicable scope, there is no difference in the configuration operations between agent-level and system-level MCP services. The following uses the process of adding and configuring an agent-level MCP service as an example to illustrate the detailed operation steps.

Add MCP service #

The Intelligent Analysis Agent provides a pre-configured system MCP service (insight-agent) that includes a variety of advanced analysis tools. You can also add more of your own MCP services.

1. Log in and enter the agent's management console. In the left navigation bar, click MCP Service Management. After entering the MCP Service Management page, click the Add MCP Service button in the upper right corner.
   

2. On the pop-up page, configure the basic information for the MCP service. After completing the configuration, click "Next".
   

   Parameters	Configuration instructions
   MCP service & description	Customize the name and description of the MCP service to identify the purpose of the added MCP service, making it easier to use and manage the MCP service later.
   Category	Select the usage type for the MCP service being added. The following options are supported: Data extension: Used to expand the data acquisition scope of the agent, typically for external data, such as online search, third-party platform data, and so on. Analysis capability: Used to enhance the agent's data analysis capabilities, such as attribution, prediction, and so on.
   Method of addition	Select the method for connecting to the MCP service. Currently supported methods are: StreamableHTTP, SSE, OpenAPI.
   URL	Configure the connection URL for the MCP service. Please ensure that the network where the DataAgent is located can access this URL.
   Header authentication	Configure the authentication information for connecting to the MCP service. If the MCP service does not require authentication, you can click the delete button next to the configuration box to remove the Header authentication configuration parameter.

3. (Optional) On the tool management page, click the Add Tool button. Configure the tool parameters in the pop-up window. Once completed, you can add the corresponding tool to the MCP service.

   tip

   When adding an MCP service via OpenAPI, tools must be added manually. When adding an MCP service via StreamableHTTP or SSE, the system will automatically retrieve the tools registered in the MCP service for you.

   

   • Basic information

     Parameters	Configuration instructions
     Tool name	Customize the name of the tool being added. The tool name will be displayed to end users on the agent usage page for intelligent analysis.
     Model invocation identifier	Configure the model invocation identifier set by the MCP service when registering the corresponding tool. warning The model invocation identifier configured here does not support Chinese characters.
     Description	Add description information for the tool to introduce its purpose to the large model, helping it invoke the tool correctly.
     Input Schema & Output Schema	Configure the format requirements for the tool's input and output parameters when the large model subsequently invokes the tool. Input schema (required): Must be configured. You need to specify the input parameter requirements for the agent when invoking the tool later. The format must be JSON Schema. Configuration example: {"type": "object", "properties": {"location": {"type": "string"}}}. Then, when the subsequent input parameter is {"location": "Seattle"} for testing, it is equivalent to executing: curl -X POST http://*****/api/v0/web_search -H 'Content-Type: application/json' -d '{"location": "Seattle"}' Output schema (optional): You can configure flexibly as needed. If you specify the format requirements for output parameters, the output result will be validated for format when the large model subsequently invokes the tool. The schema for output parameters must also be in JSON Schema format. tip After configuration is complete, for details on the validation and processing flow of input and output for MCP tool calls, see the following Reference: Processing flow when invoking MCP tools.

   • Tool functionality

     Parameters	Configuration instructions
     URL & request method	The API URL and request method for the tool must be configured.
     Request Schema & Response Schema	The Request Schema is used to convert the incoming Input Schema into the actual format accepted by the tool API. This field can be left empty. If you configure a Request Schema, the format must be in JSONata. The Response Schema is used to convert the actual format returned by the tool API into the Output Schema. This field can be left empty. If you configure a Response Schema, the format must be in JSONata. tip For detailed information about JSONata, see JSONata documentation. After configuration is complete, for details on the validation and processing flow of input and output for MCP tool calls, see the following Reference: Processing flow when invoking MCP tools.
     Header authentication	Configure the authentication information required when calling the tool. If authentication is not required, you can click the delete button next to the configuration box to remove the Header authentication configuration parameter.

4. Repeat the above steps to add multiple tools to the MCP service. After adding tools, refer to the following section to test the added tools.

Test tools #

After adding tools, you can click the "Test" button in the tool list. In the pop-up page, enter the "input parameters" for testing, and click "Test" to invoke the tool for testing.

Enable or disable MCP service #

After adding and configuring the MCP service, you can configure whether to enable or disable the MCP service on the MCP service management page, and whether to use the corresponding MCP service by default when using agents later.

When using the intelligent analysis Agent, if you select a tool from an MCP service to use, the main interaction flow between the agent's large model and the tool is as follows.

The agent's large model will follow the invocation rules of the tools in the MCP service:

• When invoking MCP tools:
  • It will validate the tool invocation input, convert the user's question to a format that complies with the MCP tool's Input schema rules, and perform validation.
  • If the MCP tool is also configured with Request schema rules, the large model will further convert the Input schema into a request format that complies with the Request schema rules and send an invocation request to the MCP tool.
• After the MCP tool completes processing and returns the invocation result:
  • If the MCP tool is configured with Response schema rules, the large model will convert the original result returned by the MCP tool into a format that complies with the Response schema rules.
  • If the MCP tool is configured with Output schema rules, the large model will further validate the returned result to check whether it complies with the Output schema rules format.