MCP 服务端实战

发表于 更新于 分类于 阅读次数：

配置环境

1
2
3
4
5
6
7
8
9
10
11
12
13
# Create a new directory for our project
uv init weather
cd weather

# Create virtual environment and activate it
uv venv
source .venv/bin/activate

# Install dependencies
uv add "mcp[cli]" httpx

# Create our server file
touch weather.py

mcp studio样例

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Math")

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

@mcp.tool()
def multiply(a: int, b: int) -> int:
    """Multiply two numbers"""
    return a * b

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

mcp streamable-http 样例

1
2
3
4
5
6
7
8
9
10
11
from mcp.server.fastmcp import FastMCP

mcp = FastMCP("Weather")

@mcp.tool()
async def get_weather(location: str) -> str:
    """Get weather for location."""
    return "It's always sunny in New York"

if __name__ == "__main__":
    mcp.run(transport="streamable-http")

使用

mcp市场MCP 广场 · 魔搭社区

使用 uv（推荐）

当使用 uv 时不需要特定的安装步骤。我们将使用 uvx 直接运行 mcp-server-fetch。

1
2
3
4
5
6
"mcpServers": {
  "fetch": {
    "command": "uvx",
    "args": ["mcp-server-fetch"]
  }
}

使用 PIP

或者，您可以通过 pip 安装 mcp-server-fetch：

1
pip install mcp-server-fetch
1
2
3
4
5
6
"mcpServers": {
  "fetch": {
    "command": "python",
    "args": ["-m", "mcp_server_fetch"]
  }
}

远程托管

image-20250802120145192 
1
2
3
4
5
6
7
8
{
  "mcpServers": {
    "fetch": {
      "type": "sse",
      "url": "https://mcp.api-inference.modelscope.net/991cf46/sse"
    }
  }
}

参考资料

构建 MCP 服务器 - 模型上下文协议 — Build an MCP Server - Model Context Protocol