第三方 MCP 伺服器

Model Context Protocol（MCP）入門

James Chapman

AI Curriculum Manager, DataCamp

為什麼用第三方 MCP 伺服器？

速度：免寫伺服器即可存取檔案系統與資料庫

為什麼用第三方 MCP 伺服器？

速度：免寫伺服器即可存取檔案系統與資料庫
他人維護：你專注於客戶端與應用程式
整合：客戶端程式碼可於不同 MCP 伺服器共用與重複使用

¹ https://www.snowflake.com/en/developers/guides/getting-started-with-snowflake-mcp-server/

使用第三方伺服器前的考量

信任：伺服器會執行程式碼並可能呼叫外部 API → 請檢查其來源
安全：需妥善管理私密資料與憑證
介面範圍：工具愈多，LLM 可做的事愈多；留意各伺服器曝露的功能

Open Library MCP

mcp-open-library：Internet Archive Open Library API 的 MCP 伺服器
工具：get_book_by_title、get_authors_by_name、get_book_cover、get_book_by_id、…
不需 API key

¹ https://github.com/8enSmith/mcp-open-library

存取 Open Library MCP

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

params = StdioServerParameters(
    command="open-library-server",
    args=[]
)

¹ https://github.com/8enSmith/mcp-open-library?tab=readme-ov-file#installation

使用 Open Library MCP

params = StdioServerParameters(command="open-library-server", args=[])


async with stdio_client(params) as (reader, writer):
    async with ClientSession(reader, writer) as session:
        await session.initialize()

        response = await session.list_tools()
        print("Tools:", [t.name for t in response.tools])
        result = await session.call_tool("get_book_by_title", {"title": "Dune"})
        text = result.content[0].text
        print("Tool Call Result:", text)

Tools: ['get_book_by_title', 'get_authors_by_name', 'get_author_info',
'get_author_photo', 'get_book_cover', 'get_book_by_id']


Tool Call Result: [
  {
    "title": "Dune",
    "authors": [
      "Frank Herbert"
    ],
    "first_publish_year": 1965,
    ...
  },
  ...

一起來練習吧！

Model Context Protocol（MCP）入門