modelcontextprotocol/python-sdk

What it is

MCP Python SDK – a Python library that implements the Model Context Protocol (MCP). MCP is a standardized way for large‑language‑model (LLM) applications to talk to external services (tools, resources, prompts) over a web‑API‑like protocol.

What you can do with it

Write MCP servers in just a few lines of Python. By decorating type‑annotated functions you expose them as tools (callable actions) or resources (templated URLs) that LLMs can invoke.
Write MCP clients that call those tools/resources over any supported transport (stdio, Streamable HTTP, Server‑Sent Events). The client handles schema generation, request/response encoding, and returns structured results.
Use the bundled CLI (mcp dev, mcp run, mcp install) for quick development, local testing, and installation of MCP packages.

How it works (high‑level)

Server side – You create an MCPServer instance, decorate functions with @mcp.tool() or @mcp.resource(). The SDK extracts the Python type hints and docstrings to build JSON‑Schema descriptions automatically, then serves them via the chosen transport.
Transport layer – MCP supports three standard transports:
- stdio – useful for embedding a server in a subprocess.
- Streamable HTTP – a lightweight HTTP endpoint that streams messages.
- SSE – Server‑Sent Events for push‑style communication.
Client side – A Client object connects to a server URL (or stdio subprocess) and provides methods like call_tool(name, args) that return a response object with structured_content.

Who might use it

LLM application developers who need a clean, typed interface for exposing custom tools (e.g., calculators, database look‑ups) to LLMs.
Platform builders creating marketplaces of reusable LLM‑compatible services.
Researchers prototyping new interaction patterns between LLMs and external code.

Getting started (quick example)

# server.py
from mcp.server import MCPServer

mcp = MCPServer("Demo")

@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two numbers."""
    return a + b

@mcp.resource("greeting://{name}")
def greeting(name: str) -> str:
    return f"Hello, {name}!"

Run it locally:

uv run mcp dev server.py   # starts a dev server (streamable‑http by default)

Client side:

import asyncio
from mcp import Client

async def main():
    async with Client("http://localhost:8000/mcp") as client:
        res = await client.call_tool("add", {"a": 1, "b": 2})
        print(res.structured_content)  # → {'result': 3}

asyncio.run(main())

Documentation & resources

Full docs: https://py.sdk.modelcontextprotocol.io/
Get‑started guide, API reference, migration guide for v1 → v2.
Protocol spec: https://modelcontextprotocol.io/specification/latest
CLI reference: mcp dev, mcp run, mcp install.

License

MIT – free to use, modify, and distribute.

All details are taken directly from the repository’s README.