MCP2Py

Install the mcp2py package using pip or uv. The uv workflow is recommended for clean dependency isolation.

pip install mcp2py
# or with uv:
uv add mcp2py

Use the load() function to launch any MCP server command and receive a Python object whose attributes are the server's tools. The server process is managed automatically.

from mcp2py import load

# Load a server by its shell command
weather = load("npx -y @h1deya/mcp-server-weather")

# Call a tool as a regular Python function
alerts = weather.get_alerts(state="CA")
print(alerts)

For HTTP(S) servers or servers requiring OAuth, pass the URL and set auth='oauth'. mcp2py handles PKCE-based browser authentication and caches tokens automatically.

from mcp2py import load

# Remote server with OAuth
notion = load("https://mcp.notion.com/mcp", auth="oauth")
me = notion.notion_get_self()
print(me)

If the MCP server requests LLM sampling (asking the client to run inference), mcp2py handles it automatically using whichever API key is available in the environment.

export ANTHROPIC_API_KEY=sk-ant-...
# or
export OPENAI_API_KEY=sk-...

For async Python applications or notebooks, use aload() to load the server without blocking the event loop.

import asyncio
from mcp2py import aload

async def main():
    server = await aload("npx async-server")
    result = await server.fetch_data(url="https://example.com")
    print(result)

asyncio.run(main())

Access the server's tool list as callable objects compatible with DSPy, Claudette, or other AI frameworks that accept function signatures.

from mcp2py import load

server = load("npx my-mcp-server")
tools = server.tools  # List of callables with type hints