Python MCP服务器示例：30分钟搭建

想把自己的数据或工具开放给Claude等AI助手。

MCP Server Example in Python: Build Your First Model Context Protocol Server in 30 Minutes

Building an MCP server in Python in 2026 is a solved problem: the official SDK (mcp) handles the protocol, and you just write plain functions. This tutorial builds a working server that exposes a tool, then connects it to Claude Desktop.

Step 1: Install

pip install "mcp[cli]"

Step 2: The Server (about 20 lines)

from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Weather Demo")

@mcp.tool()
def get_weather(city: str) -> str:
    # In production, call a real weather API here
    return "Weather in " + city + ": 22C, partly cloudy"

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

That is a complete server. One decorator, one run call.

Step 3: Test It

python server.py
# in another terminal:
mcp dev server.py

The mcp dev command opens an inspector UI where you can call get_weather and see the result.

Step 4: Connect to Claude Desktop

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "weather": {
      "command": "python",
      "args": ["C:/path/to/server.py"]
    }
  }
}

Restart Claude Desktop. Claude can now call get_weather("Tokyo") and use the result in conversation.

What the FastMCP Layer Does For You

FAQ

FastMCP or raw SDK? FastMCP for new projects - it is the official high-level API and cuts boilerplate by 80%.

Can I expose a REST API as MCP? Yes, wrap the REST calls inside a @mcp.tool() function; the LLM then calls your tool, which calls the API.

Is there a TypeScript version? Yes, the official SDK is first-class in TypeScript too. Python and TS cover the vast majority of servers.