Practical MCP with FastMCP & LangChain

Type Coercion and Validation

FastMCP validates tool arguments using Pydantic before your function runs. Pydantic's default (lax) mode performs safe type conversions when possible.

LLMs generally send correct JSON types because they receive the tool's JSON Schema, but coercion acts as a safety net. Invalid conversions fail validation and return an error before your tool runs.

To put it simply, Pydantic plays two roles in FastMCP.

• Role 1: At registration time (when @mcp.tool runs), Pydantic inspects the function signature — type hints like age: int and constraints like Field(ge=0, le=30) — and generates the JSON Schema (inputSchema