如何使用MCP开发一个客户端和服务端
一、MCP和API以及Function Call核心概念对比
| 特性 | API | Function Call | MCP (Model Context Protocol) | | 定位 | 通用应用程序接口 | 大模型原生扩展能力 | 标准化模型-服务交互协议 | | 耦合度 | 与具体服务绑定 | 与模型强绑定 (如 GPT-4-turbo) | 与模型解耦，跨平台通用 | | 交互模式 | 直接请求-响应 | 模型生成结构化调用建议 | JSON-RPC 2.0 标准化通信 | | 典型场景 | 数据集成、微服务通信 | 简单实时操作 (天气/订单查询) | 复杂异步任务 & 跨系统整合 |
二、 MCP 协议
1. 什么是MCP协议
模型上下文协议（Model Context Protocol）是一种专为大语言模型设计的标准化协议，它允许LLM以安全、一致的方式与外部系统交互。MCP协议常被描述为"AI的USB-C接口"，提供了一种统一的方式连接LLM与它们可以使用的资源。
MCP协议的核心功能包括：
- • 资源（Resources）：类似于GET端点，用于将信息加载到LLM的上下文中
- • 工具（Tools）：类似于POST端点，用于执行代码或产生副作用
- • 提示（Prompts）：可重用的LLM交互模板
- • 上下文（Context）：提供额外的交互功能，如日志记录和进度报告
2. 核心价值
- • 标准化：统一 AI 与外部服务的交互格式，解决工具碎片化问题
- • 解耦设计：模型无需硬编码 API 逻辑，通过声明式函数描述调用服务
- • 异步支持：适用于多步骤工作流（如爬取数据→分析→存储）
3. 工作流程
MCP 大概的工作方式： Claude Desktop、Cursor 这些工具，在内部实现了 MCP Client，然后MCP Client 通过标准的 MCP 协议和 MCP Server 进行交互，由各种三方开发者提供的 MCP Server 负责实现各种和三方资源交互的逻辑，比如访问数据库、浏览器、本地文件，最终再通过 标准的 MCP 协议返回给 MCP Client，最终给用户进行展示。
下图是一个通过查询天气来简单展示其对应的工作方式：- <svg aria-roledescription="sequence" role="graphics-document document" viewBox="-50 -10 1204 575" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns="http://www.w3.org/2000/svg" width="100%" id="mermaid-1749113070264"><g><rect ry="3" rx="3" name="External_Service" height="65" width="150" stroke="#666" fill="#eaeaea" y="489" x="954" ></rect><text alignment-baseline="central" dominant-baseline="central" y="521.5" x="1029"><tspan dy="0" x="1029">External_Service</tspan></text></g><g><rect ry="3" rx="3" name="MCP_Server" height="65" width="150" stroke="#666" fill="#eaeaea" y="489" x="754" ></rect><text alignment-baseline="central" dominant-baseline="central" y="521.5" x="829"><tspan dy="0" x="829">MCP_Server</tspan></text></g><g><rect ry="3" rx="3" name="AI_Model" height="65" width="150" stroke="#666" fill="#eaeaea" y="489" x="280" ></rect><text alignment-baseline="central" dominant-baseline="central" y="521.5" x="355"><tspan dy="0" x="355">AI_Model</tspan></text></g><g><rect ry="3" rx="3" name="User" height="65" width="150" stroke="#666" fill="#eaeaea" y="489" x="0" ></rect><text alignment-baseline="central" dominant-baseline="central" y="521.5" x="75"><tspan dy="0" x="75">User</tspan></text></g><g><line name="External_Service" stroke="#999" stroke-width="0.5px" y2="489" x2="1029" y1="65" x1="1029" id="actor3" ></line><g id="root-3"><rect ry="3" rx="3" name="External_Service" height="65" width="150" stroke="#666" fill="#eaeaea" y="0" x="954" ></rect><text alignment-baseline="central" dominant-baseline="central" y="32.5" x="1029"><tspan dy="0" x="1029">External_Service</tspan></text></g></g><g><line name="MCP_Server" stroke="#999" stroke-width="0.5px" y2="489" x2="829" y1="65" x1="829" id="actor2" ></line><g id="root-2"><rect ry="3" rx="3" name="MCP_Server" height="65" width="150" stroke="#666" fill="#eaeaea" y="0" x="754" ></rect><text alignment-baseline="central" dominant-baseline="central" y="32.5" x="829"><tspan dy="0" x="829">MCP_Server</tspan></text></g></g><g><line name="AI_Model" stroke="#999" stroke-width="0.5px" y2="489" x2="355" y1="65" x1="355" id="actor1" ></line><g id="root-1"><rect ry="3" rx="3" name="AI_Model" height="65" width="150" stroke="#666" fill="#eaeaea" y="0" x="280" ></rect><text alignment-baseline="central" dominant-baseline="central" y="32.5" x="355"><tspan dy="0" x="355">AI_Model</tspan></text></g></g><g><line name="User" stroke="#999" stroke-width="0.5px" y2="489" x2="75" y1="65" x1="75" id="actor0" ></line><g id="root-0"><rect ry="3" rx="3" name="User" height="65" width="150" stroke="#666" fill="#eaeaea" y="0" x="0" ></rect><text alignment-baseline="central" dominant-baseline="central" y="32.5" x="75"><tspan dy="0" x="75">User</tspan></text></g></g><g></g><defs><symbol height="24" width="24" id="computer"><path d="M2 2v13h20v-13h-20zm18 11h-16v-9h16v9zm-10.228 6l.466-1h3.524l.467 1h-4.457zm14.228 3h-24l2-6h2.104l-1.33 4h18.45l-1.297-4h2.073l2 6zm-5-10h-14v-7h14v7z" transform="scale(.5)"></path></symbol></defs><defs><symbol clip-rule="evenodd" fill-rule="evenodd" id="database"><path d="M12.258.001l.256.004.255.005.253.008.251.01.249.012.247.015.246.016.242.019.241.02.239.023.236.024.233.027.231.028.229.031.225.032.223.034.22.036.217.038.214.04.211.041.208.043.205.045.201.046.198.048.194.05.191.051.187.053.183.054.18.056.175.057.172.059.168.06.163.061.16.063.155.064.15.066.074.033.073.033.071.034.07.034.069.035.068.035.067.035.066.035.064.036.064.036.062.036.06.036.06.037.058.037.058.037.055.038.055.038.053.038.052.038.051.039.05.039.048.039.047.039.045.04.044.04.043.04.041.04.04.041.039.041.037.041.036.041.034.041.033.042.032.042.03.042.029.042.027.042.026.043.024.043.023.043.021.043.02.043.018.044.017.043.015.044.013.044.012.044.011.045.009.044.007.045.006.045.004.045.002.045.001.045v17l-.001.045-.002.045-.004.045-.006.045-.007.045-.009.044-.011.045-.012.044-.013.044-.015.044-.017.043-.018.044-.02.043-.021.043-.023.043-.024.043-.026.043-.027.042-.029.042-.03.042-.032.042-.033.042-.034.041-.036.041-.037.041-.039.041-.04.041-.041.04-.043.04-.044.04-.045.04-.047.039-.048.039-.05.039-.051.039-.052.038-.053.038-.055.038-.055.038-.058.037-.058.037-.06.037-.06.036-.062.036-.064.036-.064.036-.066.035-.067.035-.068.035-.069.035-.07.034-.071.034-.073.033-.074.033-.15.066-.155.064-.16.063-.163.061-.168.06-.172.059-.175.057-.18.056-.183.054-.187.053-.191.051-.194.05-.198.048-.201.046-.205.045-.208.043-.211.041-.214.04-.217.038-.22.036-.223.034-.225.032-.229.031-.231.028-.233.027-.236.024-.239.023-.241.02-.242.019-.246.016-.247.015-.249.012-.251.01-.253.008-.255.005-.256.004-.258.001-.258-.001-.256-.004-.255-.005-.253-.008-.251-.01-.249-.012-.247-.015-.245-.016-.243-.019-.241-.02-.238-.023-.236-.024-.234-.027-.231-.028-.228-.031-.226-.032-.223-.034-.22-.036-.217-.038-.214-.04-.211-.041-.208-.043-.204-.045-.201-.046-.198-.048-.195-.05-.19-.051-.187-.053-.184-.054-.179-.056-.176-.057-.172-.059-.167-.06-.164-.061-.159-.063-.155-.064-.151-.066-.074-.033-.072-.033-.072-.034-.07-.034-.069-.035-.068-.035-.067-.035-.066-.035-.064-.036-.063-.036-.062-.036-.061-.036-.06-.037-.058-.037-.057-.037-.056-.038-.055-.038-.053-.038-.052-.038-.051-.039-.049-.039-.049-.039-.046-.039-.046-.04-.044-.04-.043-.04-.041-.04-.04-.041-.039-.041-.037-.041-.036-.041-.034-.041-.033-.042-.032-.042-.03-.042-.029-.042-.027-.042-.026-.043-.024-.043-.023-.043-.021-.043-.02-.043-.018-.044-.017-.043-.015-.044-.013-.044-.012-.044-.011-.045-.009-.044-.007-.045-.006-.045-.004-.045-.002-.045-.001-.045v-17l.001-.045.002-.045.004-.045.006-.045.007-.045.009-.044.011-.045.012-.044.013-.044.015-.044.017-.043.018-.044.02-.043.021-.043.023-.043.024-.043.026-.043.027-.042.029-.042.03-.042.032-.042.033-.042.034-.041.036-.041.037-.041.039-.041.04-.041.041-.04.043-.04.044-.04.046-.04.046-.039.049-.039.049-.039.051-.039.052-.038.053-.038.055-.038.056-.038.057-.037.058-.037.06-.037.061-.036.062-.036.063-.036.064-.036.066-.035.067-.035.068-.035.069-.035.07-.034.072-.034.072-.033.074-.033.151-.066.155-.064.159-.063.164-.061.167-.06.172-.059.176-.057.179-.056.184-.054.187-.053.19-.051.195-.05.198-.048.201-.046.204-.045.208-.043.211-.041.214-.04.217-.038.22-.036.223-.034.226-.032.228-.031.231-.028.234-.027.236-.024.238-.023.241-.02.243-.019.245-.016.247-.015.249-.012.251-.01.253-.008.255-.005.256-.004.258-.001.258.001zm-9.258 20.499v.01l.001.021.003.021.004.022.005.021.006.022.007.022.009.023.01.022.011.023.012.023.013.023.015.023.016.024.017.023.018.024.019.024.021.024.022.025.023.024.024.025.052.049.056.05.061.051.066.051.07.051.075.051.079.052.084.052.088.052.092.052.097.052.102.051.105.052.11.052.114.051.119.051.123.051.127.05.131.05.135.05.139.048.144.049.147.047.152.047.155.047.16.045.163.045.167.043.171.043.176.041.178.041.183.039.187.039.19.037.194.035.197.035.202.033.204.031.209.03.212.029.216.027.219.025.222.024.226.021.23.02.233.018.236.016.24.015.243.012.246.01.249.008.253.005.256.004.259.001.26-.001.257-.004.254-.005.25-.008.247-.011.244-.012.241-.014.237-.016.233-.018.231-.021.226-.021.224-.024.22-.026.216-.027.212-.028.21-.031.205-.031.202-.034.198-.034.194-.036.191-.037.187-.039.183-.04.179-.04.175-.042.172-.043.168-.044.163-.045.16-.046.155-.046.152-.047.148-.048.143-.049.139-.049.136-.05.131-.05.126-.05.123-.051.118-.052.114-.051.11-.052.106-.052.101-.052.096-.052.092-.052.088-.053.083-.051.079-.052.074-.052.07-.051.065-.051.06-.051.056-.05.051-.05.023-.024.023-.025.021-.024.02-.024.019-.024.018-.024.017-.024.015-.023.014-.024.013-.023.012-.023.01-.023.01-.022.008-.022.006-.022.006-.022.004-.022.004-.021.001-.021.001-.021v-4.127l-.077.055-.08.053-.083.054-.085.053-.087.052-.09.052-.093.051-.095.05-.097.05-.1.049-.102.049-.105.048-.106.047-.109.047-.111.046-.114.045-.115.045-.118.044-.12.043-.122.042-.124.042-.126.041-.128.04-.13.04-.132.038-.134.038-.135.037-.138.037-.139.035-.142.035-.143.034-.144.033-.147.032-.148.031-.15.03-.151.03-.153.029-.154.027-.156.027-.158.026-.159.025-.161.024-.162.023-.163.022-.165.021-.166.02-.167.019-.169.018-.169.017-.171.016-.173.015-.173.014-.175.013-.175.012-.177.011-.178.01-.179.008-.179.008-.181.006-.182.005-.182.004-.184.003-.184.002h-.37l-.184-.002-.184-.003-.182-.004-.182-.005-.181-.006-.179-.008-.179-.008-.178-.01-.176-.011-.176-.012-.175-.013-.173-.014-.172-.015-.171-.016-.17-.017-.169-.018-.167-.019-.166-.02-.165-.021-.163-.022-.162-.023-.161-.024-.159-.025-.157-.026-.156-.027-.155-.027-.153-.029-.151-.03-.15-.03-.148-.031-.146-.032-.145-.033-.143-.034-.141-.035-.14-.035-.137-.037-.136-.037-.134-.038-.132-.038-.13-.04-.128-.04-.126-.041-.124-.042-.122-.042-.12-.044-.117-.043-.116-.045-.113-.045-.112-.046-.109-.047-.106-.047-.105-.048-.102-.049-.1-.049-.097-.05-.095-.05-.093-.052-.09-.051-.087-.052-.085-.053-.083-.054-.08-.054-.077-.054v4.127zm0-5.654v.011l.001.021.003.021.004.021.005.022.006.022.007.022.009.022.01.022.011.023.012.023.013.023.015.024.016.023.017.024.018.024.019.024.021.024.022.024.023.025.024.024.052.05.056.05.061.05.066.051.07.051.075.052.079.051.084.052.088.052.092.052.097.052.102.052.105.052.11.051.114.051.119.052.123.05.127.051.131.05.135.049.139.049.144.048.147.048.152.047.155.046.16.045.163.045.167.044.171.042.176.042.178.04.183.04.187.038.19.037.194.036.197.034.202.033.204.032.209.03.212.028.216.027.219.025.222.024.226.022.23.02.233.018.236.016.24.014.243.012.246.01.249.008.253.006.256.003.259.001.26-.001.257-.003.254-.006.25-.008.247-.01.244-.012.241-.015.237-.016.233-.018.231-.02.226-.022.224-.024.22-.025.216-.027.212-.029.21-.03.205-.032.202-.033.198-.035.194-.036.191-.037.187-.039.183-.039.179-.041.175-.042.172-.043.168-.044.163-.045.16-.045.155-.047.152-.047.148-.048.143-.048.139-.05.136-.049.131-.05.126-.051.123-.051.118-.051.114-.052.11-.052.106-.052.101-.052.096-.052.092-.052.088-.052.083-.052.079-.052.074-.051.07-.052.065-.051.06-.05.056-.051.051-.049.023-.025.023-.024.021-.025.02-.024.019-.024.018-.024.017-.024.015-.023.014-.023.013-.024.012-.022.01-.023.01-.023.008-.022.006-.022.006-.022.004-.021.004-.022.001-.021.001-.021v-4.139l-.077.054-.08.054-.083.054-.085.052-.087.053-.09.051-.093.051-.095.051-.097.05-.1.049-.102.049-.105.048-.106.047-.109.047-.111.046-.114.045-.115.044-.118.044-.12.044-.122.042-.124.042-.126.041-.128.04-.13.039-.132.039-.134.038-.135.037-.138.036-.139.036-.142.035-.143.033-.144.033-.147.033-.148.031-.15.03-.151.03-.153.028-.154.028-.156.027-.158.026-.159.025-.161.024-.162.023-.163.022-.165.021-.166.02-.167.019-.169.018-.169.017-.171.016-.173.015-.173.014-.175.013-.175.012-.177.011-.178.009-.179.009-.179.007-.181.007-.182.005-.182.004-.184.003-.184.002h-.37l-.184-.002-.184-.003-.182-.004-.182-.005-.181-.007-.179-.007-.179-.009-.178-.009-.176-.011-.176-.012-.175-.013-.173-.014-.172-.015-.171-.016-.17-.017-.169-.018-.167-.019-.166-.02-.165-.021-.163-.022-.162-.023-.161-.024-.159-.025-.157-.026-.156-.027-.155-.028-.153-.028-.151-.03-.15-.03-.148-.031-.146-.033-.145-.033-.143-.033-.141-.035-.14-.036-.137-.036-.136-.037-.134-.038-.132-.039-.13-.039-.128-.04-.126-.041-.124-.042-.122-.043-.12-.043-.117-.044-.116-.044-.113-.046-.112-.046-.109-.046-.106-.047-.105-.048-.102-.049-.1-.049-.097-.05-.095-.051-.093-.051-.09-.051-.087-.053-.085-.052-.083-.054-.08-.054-.077-.054v4.139zm0-5.666v.011l.001.02.003.022.004.021.005.022.006.021.007.022.009.023.01.022.011.023.012.023.013.023.015.023.016.024.017.024.018.023.019.024.021.025.022.024.023.024.024.025.052.05.056.05.061.05.066.051.07.051.075.052.079.051.084.052.088.052.092.052.097.052.102.052.105.051.11.052.114.051.119.051.123.051.127.05.131.05.135.05.139.049.144.048.147.048.152.047.155.046.16.045.163.045.167.043.171.043.176.042.178.04.183.04.187.038.19.037.194.036.197.034.202.033.204.032.209.03.212.028.216.027.219.025.222.024.226.021.23.02.233.018.236.017.24.014.243.012.246.01.249.008.253.006.256.003.259.001.26-.001.257-.003.254-.006.25-.008.247-.01.244-.013.241-.014.237-.016.233-.018.231-.02.226-.022.224-.024.22-.025.216-.027.212-.029.21-.03.205-.032.202-.033.198-.035.194-.036.191-.037.187-.039.183-.039.179-.041.175-.042.172-.043.168-.044.163-.045.16-.045.155-.047.152-.047.148-.048.143-.049.139-.049.136-.049.131-.051.126-.05.123-.051.118-.052.114-.051.11-.052.106-.052.101-.052.096-.052.092-.052.088-.052.083-.052.079-.052.074-.052.07-.051.065-.051.06-.051.056-.05.051-.049.023-.025.023-.025.021-.024.02-.024.019-.024.018-.024.017-.024.015-.023.014-.024.013-.023.012-.023.01-.022.01-.023.008-.022.006-.022.006-.022.004-.022.004-.021.001-.021.001-.021v-4.153l-.077.054-.08.054-.083.053-.085.053-.087.053-.09.051-.093.051-.095.051-.097.05-.1.049-.102.048-.105.048-.106.048-.109.046-.111.046-.114.046-.115.044-.118.044-.12.043-.122.043-.124.042-.126.041-.128.04-.13.039-.132.039-.134.038-.135.037-.138.036-.139.036-.142.034-.143.034-.144.033-.147.032-.148.032-.15.03-.151.03-.153.028-.154.028-.156.027-.158.026-.159.024-.161.024-.162.023-.163.023-.165.021-.166.02-.167.019-.169.018-.169.017-.171.016-.173.015-.173.014-.175.013-.175.012-.177.01-.178.01-.179.009-.179.007-.181.006-.182.006-.182.004-.184.003-.184.001-.185.001-.185-.001-.184-.001-.184-.003-.182-.004-.182-.006-.181-.006-.179-.007-.179-.009-.178-.01-.176-.01-.176-.012-.175-.013-.173-.014-.172-.015-.171-.016-.17-.017-.169-.018-.167-.019-.166-.02-.165-.021-.163-.023-.162-.023-.161-.024-.159-.024-.157-.026-.156-.027-.155-.028-.153-.028-.151-.03-.15-.03-.148-.032-.146-.032-.145-.033-.143-.034-.141-.034-.14-.036-.137-.036-.136-.037-.134-.038-.132-.039-.13-.039-.128-.041-.126-.041-.124-.041-.122-.043-.12-.043-.117-.044-.116-.044-.113-.046-.112-.046-.109-.046-.106-.048-.105-.048-.102-.048-.1-.05-.097-.049-.095-.051-.093-.051-.09-.052-.087-.052-.085-.053-.083-.053-.08-.054-.077-.054v4.153zm8.74-8.179l-.257.004-.254.005-.25.008-.247.011-.244.012-.241.014-.237.016-.233.018-.231.021-.226.022-.224.023-.22.026-.216.027-.212.028-.21.031-.205.032-.202.033-.198.034-.194.036-.191.038-.187.038-.183.04-.179.041-.175.042-.172.043-.168.043-.163.045-.16.046-.155.046-.152.048-.148.048-.143.048-.139.049-.136.05-.131.05-.126.051-.123.051-.118.051-.114.052-.11.052-.106.052-.101.052-.096.052-.092.052-.088.052-.083.052-.079.052-.074.051-.07.052-.065.051-.06.05-.056.05-.051.05-.023.025-.023.024-.021.024-.02.025-.019.024-.018.024-.017.023-.015.024-.014.023-.013.023-.012.023-.01.023-.01.022-.008.022-.006.023-.006.021-.004.022-.004.021-.001.021-.001.021.001.021.001.021.004.021.004.022.006.021.006.023.008.022.01.022.01.023.012.023.013.023.014.023.015.024.017.023.018.024.019.024.02.025.021.024.023.024.023.025.051.05.056.05.06.05.065.051.07.052.074.051.079.052.083.052.088.052.092.052.096.052.101.052.106.052.11.052.114.052.118.051.123.051.126.051.131.05.136.05.139.049.143.048.148.048.152.048.155.046.16.046.163.045.168.043.172.043.175.042.179.041.183.04.187.038.191.038.194.036.198.034.202.033.205.032.21.031.212.028.216.027.22.026.224.023.226.022.231.021.233.018.237.016.241.014.244.012.247.011.25.008.254.005.257.004.26.001.26-.001.257-.004.254-.005.25-.008.247-.011.244-.012.241-.014.237-.016.233-.018.231-.021.226-.022.224-.023.22-.026.216-.027.212-.028.21-.031.205-.032.202-.033.198-.034.194-.036.191-.038.187-.038.183-.04.179-.041.175-.042.172-.043.168-.043.163-.045.16-.046.155-.046.152-.048.148-.048.143-.048.139-.049.136-.05.131-.05.126-.051.123-.051.118-.051.114-.052.11-.052.106-.052.101-.052.096-.052.092-.052.088-.052.083-.052.079-.052.074-.051.07-.052.065-.051.06-.05.056-.05.051-.05.023-.025.023-.024.021-.024.02-.025.019-.024.018-.024.017-.023.015-.024.014-.023.013-.023.012-.023.01-.023.01-.022.008-.022.006-.023.006-.021.004-.022.004-.021.001-.021.001-.021-.001-.021-.001-.021-.004-.021-.004-.022-.006-.021-.006-.023-.008-.022-.01-.022-.01-.023-.012-.023-.013-.023-.014-.023-.015-.024-.017-.023-.018-.024-.019-.024-.02-.025-.021-.024-.023-.024-.023-.025-.051-.05-.056-.05-.06-.05-.065-.051-.07-.052-.074-.051-.079-.052-.083-.052-.088-.052-.092-.052-.096-.052-.101-.052-.106-.052-.11-.052-.114-.052-.118-.051-.123-.051-.126-.051-.131-.05-.136-.05-.139-.049-.143-.048-.148-.048-.152-.048-.155-.046-.16-.046-.163-.045-.168-.043-.172-.043-.175-.042-.179-.041-.183-.04-.187-.038-.191-.038-.194-.036-.198-.034-.202-.033-.205-.032-.21-.031-.212-.028-.216-.027-.22-.026-.224-.023-.226-.022-.231-.021-.233-.018-.237-.016-.241-.014-.244-.012-.247-.011-.25-.008-.254-.005-.257-.004-.26-.001-.26.001z" transform="scale(.5)"></path></symbol></defs><defs><symbol height="24" width="24" id="clock"><path d="M12 2c5.514 0 10 4.486 10 10s-4.486 10-10 10-10-4.486-10-10 4.486-10 10-10zm0-2c-6.627 0-12 5.373-12 12s5.373 12 12 12 12-5.373 12-12-5.373-12-12-12zm5.848 12.459c.202.038.202.333.001.372-1.907.361-6.045 1.111-6.547 1.111-.719 0-1.301-.582-1.301-1.301 0-.512.77-5.447 1.125-7.445.034-.192.312-.181.343.014l.985 6.238 5.394 1.011z" transform="scale(.5)"></path></symbol></defs><defs><marker orient="auto-start-reverse" markerHeight="12" markerWidth="12" markerUnits="userSpaceOnUse" refY="5" refX="7.9" id="arrowhead"><path d="M -1 0 L 10 5 L 0 10 z" ></path></marker></defs><defs><marker refY="4.5" refX="4" orient="auto" markerHeight="8" markerWidth="15" id="crosshead"><path d="M 1,2 L 6,7 M 6,2 L 1,7" stroke-width="1pt" stroke="#000000" fill="none"></path></marker></defs><defs><marker orient="auto" markerHeight="28" markerWidth="20" refY="7" refX="15.5" id="filled-head"><path d="M 18,7 L9,13 L14,7 L9,1 Z"></path></marker></defs><defs><marker orient="auto" markerHeight="40" markerWidth="60" refY="15" refX="15" id="sequencenumber" ><circle r="6" cy="15" cx="15"></circle></marker></defs><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="80" x="214">“查询北京天气”</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="123" x2="351" y1="123" x1="76"></line><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="138" x="356">解析意图，生成MCP调用</text><path marker-end="url(#arrowhead)" stroke="none" stroke-width="2" d="M 356,173 C 416,163 416,203 356,193"></path><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="218" x="591">{ "function": "get_weather", "params": {"city":"北京"} }</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="261" x2="825" y1="261" x1="356"></line><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="276" x="928">调用天气API</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="311" x2="1025" y1="311" x1="830"></line><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="326" x="931">返回原始数据</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="361" x2="833" y1="361" x1="1028"></line><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="376" x="594">{ "temperature": "22°C", "condition": "晴" }</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="411" x2="359" y1="411" x1="828"></line><text dy="1em" alignment-baseline="middle" dominant-baseline="middle" text-anchor="middle" y="426" x="217">“北京今天晴天，气温22°C”</text><line marker-end="url(#arrowhead)" stroke="none" stroke-width="2" y2="469" x2="79" y1="469" x1="354"></line></svg>
现在python编写mcp server和mcp client的有两个分别是FastMCP 和MCP，其中MCP是官方的pythonsdk,这两个之间的关系是官方收编了FastMCP的第一个版本的包，但官方集成的是 fastmcp 的 v1.0 版本。然而，jlowin 继续开发 fastmcp，还发布了 v2.0 版本，其中包含代理和客户端采样等新功能。以下的演示以官方版本MCP为例，
安装：uv add "mcp[cli]” 或者pip install "mcp[cli]”
(1) MCP 服务端
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg>from mcp.server.fastmcp import FastMCP
- from mcp.server.fastmcp.prompts import base
- # mcp = FastMCP(name="demo",host="127.0.0.1",port=8256,sse_path="/sse") ### 启动方式为sse时使用
- mcp = FastMCP()
- @mcp.tool()
- def add_2_numbers(a: int, b: int) -> int:
- """两个数字相加"""
- return a + b
- @mcp.resource("config://app")
- def get_config() -> str:
- """Static configuration data"""
- return "App configuration here"
- @mcp.prompt()
- def debug_error(error: str) -> list[base.Message]:
- return [
- base.UserMessage("I'm seeing this error:"),
- base.UserMessage(error),
- base.AssistantMessage("I'll help debug that. What have you tried so far?"),
- @mcp.tool()
- def multiply_2_numbers(a: int, b: int):
- """两个数字相乘"""
- return a * b
- if __name__ == "__main__":
- # mcp.run(transport='sse') ## 启动方式为sse
- mcp.run(transport='stdio') ## 启动方式为stdio
- • Tools（工具）是MCP中最常用的功能之一，它允许LLM执行特定的操作或函数。使用@mcp.tool()装饰器可以轻松将Python函数转换为LLM可调用的工具：
- • Resources（资源）用于向LLM提供数据和上下文信息。与工具不同，资源主要用于读取数据而非执行操作
- • Prompts（提示）允许您创建可重用的提示模板，这些模板可以被参数化并用于标准化LLM交互
简单验证服务端功能可以通过mcp dev server.py进入界面检测
(2) MCP 客户端
MCP客户端一般分别按照服务端的stdio和sse分别写了两个，具体融合的最后修改一下即可。
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg>import asyncio
- import json
- import re
- from contextlib import AsyncExitStack
- from typing import Optional
- from lxml import etree
- from mcp import ClientSession, StdioServerParameters, stdio_client
- from mcp.client.sse import sse_client
- from openai import AsyncOpenAI
- class Stdio_MCPClient():
- def __init__(self,api_key, base_url, model):
- self.session: Optional[ClientSession] = None
- self.exit_stack = AsyncExitStack()
- self.client = AsyncOpenAI(api_key=api_key, base_url=base_url)
- self.model = model
- self.message = []
- with open("MCP_Prompt.txt", "r", encoding="utf-8") as f:
- self.system_prompt = f.read()
- async def connect_to_stdio_server(self, mcp_name, command,args,env={}):
- server_params = StdioServerParameters(
- command=command,
- args=args,
- env=env
- )
- stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
- self.stdio,self.write = stdio_transport
- self.session = await self.exit_stack.enter_async_context(ClientSession(self.stdio, self.write))
- await self.session.initialize()
- response = await self.session.list_tools()
- tools = response.tools
- print(f"成功链接到{mcp_name}服务，对应的tools：",[tool.name for tool in tools])
- self.available_tools = [{
- "type": "function",
- "function": {
- "name": tool.name,
- "description": tool.description,
- "parameters": tool.inputSchema
- }
- } for tool in response.tools]
- async def process_query(self, query: str, stream: bool = False):
- # self.message.append({"role": "system", "content": self.system_prompt})
- self.message.append({"role": "user", "content": query})
- response = await self.client.chat.completions.create(
- model=self.model,
- messages=self.message,
- tools=self.available_tools
- )
- final_text = []
- assistant_message = response.choices[0].message
- while assistant_message.tool_calls:
- for tool_call in assistant_message.tool_calls:
- tool_name = tool_call.function.name
- tool_args = json.loads(tool_call.function.arguments)
- result = await self.session.call_tool(tool_name, tool_args)
- print(f"calling tools {tool_name},wirh args {tool_args}")
- print("Result:", result.content[0].text)
- self.message.extend([
- {
- "role": "assistant",
- "content": None,
- "tool_calls": [tool_call]
- },
- {
- "role": "tool",
- "content": result.content[0].text,
- "tool_call_id": tool_call.id
- }
- ])
- response = await self.client.chat.completions.create(
- model=self.model,
- messages=self.message,
- tools=self.available_tools,
- max_tokens=8048
- )
- assistant_message = response.choices[0].message
- content = assistant_message.content
- final_text.append(content)
- return "\n".join(final_text)
- async def chat_loop(self,stream_mode=True):
- self.message = []
- while True:
- try:
- query = input("\nQuery: ").strip()
- if query.lower() == 'quit':
- break
- if query.strip() == '':
- continue
- response = await self.process_query(query, stream=stream_mode)
- print("\nAI:", response)
- except Exception as e:
- print(f"\nError: {str(e)}")
- async def cleanup(self):
- await self.exit_stack.aclose()
- async def main():
- with open("config.json", "r") as f:
- config = json.load(f)
- client = Stdio_MCPClient(config["llm"]["api_key"], config["llm"]["base_url"], config["llm"]["model"])
- try:
- env = {}
- await client.connect_to_stdio_server("testserver","python",["server.py",],{})
- await client.chat_loop()
- except Exception as e:
- print(e)
- finally:
- await client.cleanup()
- if __name__ == '__main__':
- asyncio.run(main())
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg>import asyncio
- import json
- import re
- from contextlib import AsyncExitStack
- from typing import Optional
- from lxml import etree
- from mcp import ClientSession, StdioServerParameters, stdio_client
- from mcp.client.sse import sse_client
- from openai import AsyncOpenAI
- class SSE_MCPClient():
- def __init__(self,api_key, base_url, model):
- self.session: Optional[ClientSession] = None
- self.exit_stack = AsyncExitStack()
- self.client = AsyncOpenAI(api_key=api_key, base_url=base_url)
- self.model = model
- self.message = []
- with open("MCP_Prompt.txt", "r", encoding="utf-8") as f:
- self.system_prompt = f.read()
- async def connect_to_sse_server(self, mcp_name, server_url,headers=None):
- self.service = [{
- "name": mcp_name,
- "url": server_url,
- "headers": headers
- }]
- sse_transport = await self.exit_stack.enter_async_context(sse_client(server_url, headers,timeout=30,sse_read_timeout=30))
- self.sse,self.write = sse_transport
- self.session = await self.exit_stack.enter_async_context(ClientSession(self.sse, self.write))
- await self.session.initialize()
- response = await self.session.list_tools()
- tools = response.tools
- print(f"成功链接到{mcp_name}服务，对应的tools：",[tool.name for tool in tools])
- self.available_tools = [{
- "type": "function",
- "function": {
- "name": tool.name,
- "description": tool.description,
- "parameters": tool.inputSchema
- }
- } for tool in response.tools]
- async def reconnect_sse_server(self):
- for service in self.service:
- mcp_name = service["name"]
- server_url = service["url"]
- headers = service.get("headers", None)
- sse_transport = await self.exit_stack.enter_async_context(sse_client(server_url, headers))
- self.sse, self.write = sse_transport
- self.session = await self.exit_stack.enter_async_context(ClientSession(self.sse, self.write))
- await self.session.initialize()
- print(f"重新成功链接到 {mcp_name} 服务")
-
- async def process_query(self, query: str, stream: bool = False):
- # self.message.append({"role": "system", "content": self.system_prompt})
- self.message.append({"role": "user", "content": query})
- response = await self.client.chat.completions.create(
- model=self.model,
- messages=self.message,
- tools=self.available_tools
- )
- final_text = []
- assistant_message = response.choices[0].message
- while assistant_message.tool_calls:
- for tool_call in assistant_message.tool_calls:
- tool_name = tool_call.function.name
- tool_args = json.loads(tool_call.function.arguments)
- await self.reconnect_sse_server()
- result = await self.session.call_tool(tool_name, tool_args)
- print(f"calling tools {tool_name},wirh args {tool_args}")
- print("Result:", result.content[0].text)
- self.message.extend([
- {
- "role": "assistant",
- "content": None,
- "tool_calls": [tool_call]
- },
- {
- "role": "tool",
- "content": result.content[0].text,
- "tool_call_id": tool_call.id
- }
- ])
- response = await self.client.chat.completions.create(
- model=self.model,
- messages=self.message,
- tools=self.available_tools,
- max_tokens=8048
- )
- assistant_message = response.choices[0].message
- content = assistant_message.content
- final_text.append(content)
- return "\n".join(final_text)
- async def chat_loop(self,stream_mode=True):
- self.message = []
- while True:
- try:
- query = input("\nQuery: ").strip()
- if query.lower() == 'quit':
- break
- if query.strip() == '':
- continue
- response = await self.process_query(query, stream=stream_mode)
- print("\nAI:", response)
- except Exception as e:
- print(f"\nError: {str(e)}")
- async def cleanup(self):
- await self.exit_stack.aclose()
- async def main():
- with open("config.json", "r") as f:
- config = json.load(f)
- client = Stdio_MCPClient(config["llm"]["api_key"], config["llm"]["base_url"], config["llm"]["model"])
- try:
- await client.connect_to_sse_server(mcp_name="test", server_url="https://127.0.0.1:7860/sse")
- await client.chat_loop()
- except Exception as e:
- print(e)
- finally:
- await client.cleanup()
- if __name__ == '__main__':
- asyncio.run(main())
（3）版本的自然更新
有了上面两种客户端的连接方法，自然而然结合两个就可以做到同时结合sse和stdio的方法只需要增加一个分别调用的方法即可，后续代码微微改动便可使用。
当然官方的MCP也是在不段更新的，看了官方有发布Streamable HTTP Transport ,这种方式在取代sse,以及通过with来启动执行服务的更新等等，一些简单的更新参考下面，更多更新可以前往github上看- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg>from mcp.server.fastmcp import FastMCP
- # Stateful server (maintains session state)
- mcp = FastMCP("StatefulServer")
- # Stateless server (no session persistence)
- mcp = FastMCP("StatelessServer", stateless_http=True)
- # Stateless server (no session persistence, no sse stream with supported client)
- mcp = FastMCP("StatelessServer", stateless_http=True, json_response=True)
- # Run server with streamable_http transport
- mcp.run(transport="streamable-http")
三、典型应用场景
1. MCP 适用场景
- • 企业系统整合 将 CRM/ERP 封装为 MCP 服务，供多个 Agent 安全调用
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg># MCP 连接数据库示例
- @app.post("/mcp")
- def query_database(request: dict):
- if request["function"] == "get_user_orders":
- user_id = request["parameters"]["user_id"]
- # 执行SQL查询 (伪代码)
- return {"orders": db.query(f"SELECT * FROM orders WHERE user_id={user_id}")}
- • 跨平台自动化 组合 GitHub + Slack 的 MCP 服务实现 CI/CD 流程：
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg># 自动化工作流：提交代码→构建→通知
- def ci_cd_pipeline():
- call_mcp("github", {"action": "pull_code", "repo": "my-app"})
- build_result = call_mcp("jenkins", {"job": "build"})
- call_mcp("slack", {"channel": "dev-team", "message": f"构建结果：{build_result}"})
2. Function Call 适用场景
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg># 简单实时查询（无需MCP）
- def get_stock_price(symbol: str):
- return yahoo_finance_api(symbol)
- # 注册函数到模型
- functions = [{
- "name": "get_stock_price",
- "parameters": {"symbol": {"type": "string"}}
- }]
- # 模型直接调用
- response = model.generate("AAPL当前股价？", functions=functions)
- if response.function_call:
- print(get_stock_price(response.function_call.arguments["symbol"]))
- <svg xmlns="http://www.w3.org/2000/svg" version="1.1" x="0px" y="0px" width="45px" height="13px" viewBox="0 0 450 130">
- <ellipse cx="50" cy="65" rx="50" ry="52" stroke="rgb(220,60,54)" stroke-width="2" fill="rgb(237,108,96)"></ellipse>
- <ellipse cx="225" cy="65" rx="50" ry="52" stroke="rgb(218,151,33)" stroke-width="2" fill="rgb(247,193,81)"></ellipse>
- <ellipse cx="400" cy="65" rx="50" ry="52" stroke="rgb(27,161,37)" stroke-width="2" fill="rgb(100,200,86)"></ellipse>
- </svg># 直接调用 REST API（无AI参与）
- import requests
- def fetch_weather(city: str):
- response = requests.get(f"https://api.weather.com/v1/{city}")
- return response.json()["temperature"]
四、技术选型建议
| 场景 | 推荐方案 | 原因 | | 简单同步任务（天气/股价查询） | Function Call | 低延迟，与模型紧密集成 | | 跨系统异步任务（数据分析流水线） | MCP | 标准化协议支持复杂工作流 | | 企业内部系统暴露服务 | MCP | 统一认证 + 访问控制 | | 第三方公共服务调用 | API + Function Call | 无需额外协议层 |
关键结论：MCP 的核心价值在于建立企业级 AI 基础设施。当系统需要连接多个异构数据源、要求严格的协议标准化或涉及长周期任务时，MCP 是优于 Function Call 的选择。
来源：程序园用户自行投稿发布，如果侵权，请联系站长删除
免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！ |
提升卡
置顶卡
沉默卡
喧嚣卡
变色卡
千斤顶
照妖镜
