add-mcp-tool

To expose a Python helper as an MCP tool, you wrap it in a FastMCP tool registration with Pydantic-typed request and response models. This enforces a tool→service→helper call chain, ensuring AI clients can safely call the function with deterministic, validation-ready schemas.

Pydantic typing in FastMCP tool generation defines strict request and response models for your helper functions. This produces deterministic, validation-ready tool interfaces that AI hosts can reliably parse, preventing schema drift when AI assistants call your service layer operations.

The best way to test MCP tools without real HTTP calls is to mock the client at the service boundary. You then assert tool outputs using unwrap_tool_result for clean typed verification, ensuring tests remain isolated and validate the tool→service→helper chain without external dependencies.

FastMCP tool generation requires a layered service architecture where tools call services and services call helpers. This design prevents layer bypasses, keeps behavior testable and maintainable, and ensures results return through make_json_result to maintain consistent MCP integration without content-channel drift.

MCP tool output drifts from expected JSON formats when not using make_json_result. This helper function standardizes JSON output, preventing content-channel drift and keeping MCP integration consistent across all tool registrations exposed to AI clients.

Yes, you can expose an existing OpenAPI client helper as an MCP tool. The process creates a FastMCP tool registration with Pydantic-typed I/O models, wrapping the helper in a service layer to enforce the required tool→service→helper call chain for safe AI client access.