Different tools for different paths (similar to routes)

[am-akash1309]

I been wondering if I can have the tools grouped under different routes.

mcp.run( transport="http", port=8000, log_level="debug", path="/db_tools", )

I have noticed that all the tools from the object mcp can be accessed through /db_tools.
So, let's say I have different agent for reading and writing in a database. Will it be possible to create two different paths in the same port as /db_write and /db_read? So that I can access only the particular tools that an agent needs and not all in the server.

[KaliszS]

Hi @am-akash1309! run() method from FastMCP instance doesn't support path parameter. You can use this as a flag for start command (fastmcp run <server_module>.py --path="/db_tools") though, but it will only replace /mcp endpoint name.

You can have tools separated in form of different modules (example: https://github.com/the-momentum/apple-health-mcp-server/tree/main/app/mcp/v1/tools), with each of them being defined with its own FastMCP server instance. Then you can merge them all into one MCP Server in the main module (using mount() method), but one MCP server is just a collection of tools, without grouping them.

If you want to split tools for agentic usage, the best way to go is probably to have separate MCP Servers. Here is an example of pydantic ai agents with MCP integration: https://ai.pydantic.dev/mcp/client/#tool-call-customisation. You just register each MCP server as a specific toolset.