MCP Server Setup and Testing

English Summary

This tutorial explains how to set up and test a simple MCP server:

Installation

• Install MCP CLI tools: pip install mcp
• Install UV (modern packaging and execution tool): pip install uv
• Node.js is also required for the inspector interface

Development Process

1. Create a file named server.py
2. Import the MCP module: from mcp.server.fetch import mcp
3. Create an MCP server instance: mcp = fetch.mcp("demo")
4. Define tools using the @mcp.tool decorator
5. Implement functions like add or divide
6. Add mcp.run() to start the server

Testing

• Run the development server with: mcp dev server.py
• This opens an inspector web interface at localhost:5173
• Click "Connect" in the inspector to establish a connection
• Test tools by providing parameters and checking responses
• Handle errors by using try/except and return error messages with the "error" key

Deployment

• For production environments, use: mcp connect run server.py
• Development mode includes auto-reload and debugging features

# server.py

# 匯入 FastMCP 類別，用來建立 MCP 伺服器實例
from mcp.server.fastmcp import FastMCP

# 建立一個 MCP 伺服器實例，名稱為 "Demo"
# 這個實例會管理所有定義的工具（tool）與資源（resource）
mcp = FastMCP("Demo")


# ========================
# 定義一個「工具」Tool 範例
# ========================

# 使用 @mcp.tool() 裝飾器，把這個函式註冊為 MCP 的工具
# 工具可以被遠端呼叫，例如從 HTTP API 請求中觸發
@mcp.tool()
def add(a: int, b: int) -> int:
    """
    將兩個整數相加，回傳總和。
    這是一個最簡單的 MCP 工具示範。
    """
    try:
        result = a + b
        return {"result":result}
    except Exception as e:
        return {"error":str(e)}


# 啟動 MCP HTTP 伺服器
if __name__ == "__main__":
    mcp.run()