FastMCP

What it does

• Decorator-driven API: @mcp.tool / @mcp.resource / @mcp.prompt — your Python function becomes a server primitive
• Type-hint schemas: JSON Schema auto-generated from Python type annotations, including Pydantic models
• Transports: stdio for local clients (Claude Code, Inspector), HTTP / Streamable HTTP for remote
• Built-in helpers: image / audio / PDF content, sampling, elicitation, structured prompts
• Server composition: mount / import one FastMCP server inside another; auth providers (Google, GitHub, Azure)

When to use it

Best for shipping an MCP server in an afternoon — wrap an existing internal API as MCP, expose a database as queryable resources, or build the custom tool side of an Agent SDK project. The de-facto Python choice; pair with uv for fast setup.

Getting started

uv add fastmcp

Then:

from fastmcp import FastMCP
mcp = FastMCP("my-server")
@mcp.tool
def add(a: int, b: int) -> int: return a + b
mcp.run()

Run via python -m my_server for stdio, or pass transport="http" for the Streamable HTTP server.

Things to know

FastMCP has its own major-version cadence (currently v2.x) — pin it in production. The async-first API works best with uvicorn-style event loops; sync tool functions are supported but the framework encourages async def.