Build an MCP server in Python (FastMCP tutorial, 2026)

Building an MCP server in Python is short work thanks to FastMCP, the high-level helper in the official Python SDK. Here is a minimal server that exposes one tool — under 30 lines — runnable in Claude Desktop and Cursor.

Setup

Use uv (recommended) or pip:

uv init weather-mcp && cd weather-mcp
uv add "mcp[cli]"

The server

Create server.py:

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("weather")

@mcp.tool()
def get_weather(city: str) -> str:
    """Get the current weather for a given city."""
    # TODO: call a real weather API. For now, fake it.
    return f"{city}: 18°C, partly cloudy"

if __name__ == "__main__":
    mcp.run()

That is the whole server. The @mcp.tool() decorator turns a typed function into an MCP tool — the docstring becomes the tool description, and type hints become the input schema. No boilerplate.

Test it

Run the MCP inspector against it:

uv run mcp dev server.py

This opens a browser UI. List tools → you should see get_weather. Call it with a city and you get the fake forecast.

Use it in Claude Desktop

Add to your config (see MCP config file location):

{
  "mcpServers": {
    "weather": {
      "command": "uv",
      "args": ["--directory", "/absolute/path/to/weather-mcp", "run", "server.py"]
    }
  }
}

Restart Claude, click the hammer icon, and ask "what's the weather in Tokyo?" — it calls your tool.

Make it real

• Replace the fake body with a real HTTP call (httpx). Read the API key from an environment variable.
• Add more tools — each is another @mcp.tool() function.
• Log to stderr, never stdout — stdout carries the MCP protocol, and printing there closes the connection. See MCP "connection closed" error.

Going further

Prefer TypeScript? See build your first MCP server in Node.js. Already have a REST API? Generate an MCP server from OpenAPI. When it is ready, submit it to Loadout.