什么是MCP?---**Title:** MCP AI时代的万物互联 **Section Title:** 未来展望 **Subsection Title:** 服务端 **Paragraph:** MCP 正在打造开放生态, 不少公司已经在提供 MCP 服务, 就像 App Store 一样热闹! 宝贝们, 这波风口咱们必须赶上啊! MCP 协议极有可能 dominate, 接口商 业化也是极有可能的, 这一点利好卖接口服务 的中间商。 **Section Title:** TMCP服务器列表: **List:** * 21st.dev Magic 灵感驱动 UI 组件 (21st 设计团队) * Apify 3000 + 云爬虫 (电商 / 社交 / 地图数据) * Cloudflare 云部署、配置和查询您的资源 * Exa AI 专用搜索引擎 * Firecrawl 网页数据提取 * Kagi Search 网络搜索 (API 驱动) * ... * 【留言+私信】服务器连接【即可获得访问权限】 **Button/Annotation:** 建议收藏 MCP AI时代的万物互联 未来展望 Manus的火爆验证了通用型agent的市场需求，也加速了整个行业的资源投入和技术迭代（比如MCP协议）。 有意思的是，Manus联合创始人表示Manus并未采用MCP协议，而是借鉴了CodeAct框架：统一Agent行动为可执行的Python代码，以提升其交互能力和处理复杂任务的能力。 预计Manus采用MCP协议以后manus性能会有一步的提升，比如不用单独开沙盒虚拟机，因为服务提供商能直接设定agent的权限。 客户端 客户端主要是AI模型公司，Agent公司： - 首先，主流大型模型普遍缺乏对MCP协议的深度兼容 - 其次，通用Agent要求模型严格遵循预设流程执行任务。Manus和OWL这些Agent产品，模型不会记住每次成功完成任务的路径，每一次执行都是从头开始：一次配置的环境不成功再重复配一次，一个工具使用了就不调用另一个工具。这种强制顺序执行的特性与限定了模型需要按照规划一步一步执行下去，导致算力成本大幅上升。 - 理论上，MCP的工具可以无限扩展，即使给模型提供有文字描述功能的模块，大型模型本身不可能把所有的工具内容都训练进去。 因此，未来模型应当学习如何更好地使用MCP，比如在执行任务的过程中收集反馈，保存成功执行的路径并运用在以后类似的任务中。未来模型如果能记住最优小代价的成功路径，性能还能再进一步提升。 建议收藏 **Title:** MCP AI时代的万物互联 (MCP Internet of Everything in the AI Era) **Subtitle:** 三种场景 (Three Scenarios) **Scenarios:** 1. 旅行规划助手 (Travel Planning Assistant) * a. 使用 API 时：分别为谷歌日历、邮件、机票预订写代码，繁琐而复杂。 (When using API: Need to write code for Google Calendar, email, flight booking separately, which is tedious and complex.) * b. 使用 MCP 时：AI助手Agent直接通过MCP统一协议，查阅日历、订机票、发邮件确认，无须单独整合每个工具。 (When using MCP: AI Assistant Agent directly checks calendar, books flights, sends email confirmations through the unified MCP protocol, without needing to integrate each tool separately.) 2. 智能 IDE (代码编辑器) (Smart IDE (Code Editor)) * a. 使用 API 时：手动连接文件系统、版本管理、包管理和文档，耗时费力。 (When using API: Manually connect file system, version control, package management, and documentation, which is time-consuming and laborious.) * b. 使用 MCP 时：IDE 通过 MCP 一次连接所有功能，带来更丰富的上下文支持，更强大的智能建议。 (When using MCP: IDE connects all functions at once through MCP, bringing richer context support and more powerful smart suggestions.) 3. 复杂的数据分析 (Complex Data Analysis) * a. 使用 API 时：人为管理与每个数据库、数据可视化工具的连接。 (When using API: Manually manage the connection to each database and data visualization tool.) * b. 使用 MCP 时：Agent 自动发现并连接多个数据库和可视化工具，通过统一的MCP接口完成分析任务。 (When using MCP: Agent automatically discovers and connects multiple databases and visualization tools, completing analysis tasks through the unified MCP interface.) **Annotation:** 建议收藏 (Recommended for Collection) Here is the extraction from the image: **Title:** MCP AI时代的万物互联 **Question:** 💡什么是MCP? **Diagram:** * **Type:** Schematic diagram illustrating system architecture and connectivity. * **Title:** MCP architecture * **Main Elements:** * A laptop representing the core system/client. * Box labeled "MCP clients" with "client.py" below it, positioned near the laptop screen. An arrow points from the laptop towards this box. Another dashed arrow points from "MCP clients" to the area above "MCP hosts". * Box labeled "MCP hosts" with symbols for Claude, a swirling logo (possibly Anthropic's logo), and hexagons below it. An arrow points from the area above "MCP hosts" towards "MCP hosts". * Box labeled "MCP" positioned near the laptop's trackpad area. * Three boxes labeled "MCP server" positioned below the laptop and connected to various external devices/services via cables depicted as USB-like connections. * Box labeled "Remote services" with icons representing various applications (e.g., colored dots, "M" logo, calendar icon with "31"). This box is connected via a cable to one "MCP server". * Box labeled "Local data sources" with a folder icon and an Apple Finder-like icon. This box is connected via a USB cable to another "MCP server" via the laptop's USB port. * **Connections:** Cables and arrows show interactions and connections between components, suggesting the MCP server acts as an intermediary between the core system/clients and external services/data sources. Dashed arrows suggest communication flows. **Explanation Text:** 简单来说, MCP 定了个统一标准, 不管是什么Agent或者AI助手 (Windsurf、Cursor还是Cladue), 只要符合这个协议, 就能连接到后端或不同 API, 直接一个 MCP 协议搞定, 就像是为AI量身定制的“USB-C”⚡。 **Section Title:** 🔥核心架构大揭秘 **Core Architecture Details:** ✅MCP客户端: AI模型内部使用MCP协议与MCP服务器通信。 ✅MCP服务器: 连接 AI 模型和外部系统之间的中介, 充当“桥梁”和“翻译”。它接收来自 MCP 客户端的请求, 理解 AI 的意图, 然后去外部资源获取数据或执行操作, 再把结果返回给 AI 模型。 **Final Note:** 建议收藏 **Title:** MCP\ AI时代的万物互联 **Question:** 什么是 MCP? **Table:** | 功能 | 传统API | MCP协议 | | :----------------- | :-------------- | :------------ | | 整合难度 | 每个API单独整合 | 一次标准化整合 | | 实时双向通信 | ❌ 不支持 | ✅ 支持 | | 动态发现工具 | ❌ 不支持 | ✅ 支持 | | 扩展性 | (弱) 需要额外开发 | (强) 即插即用 | | 安全性与控制 | 每个API单独定义 | 所有工具统一标准 | **Other Text:** # AI 界的USB-C # 建议收藏 **Diagram Description:** * **Type:** Component Diagram / Flowchart * **Overall Structure:** The diagram shows the relationships between several key components of a system, visually arranged in a hierarchical manner. There are two main components at the top, connected to several sub-components below. * **Title:** "Key Components" is centered at the top. * **Main Components (Top Level):** * Left: A rounded rectangle labeled "MCP Client" with an icon of a globe overlaid by a cursor. * Right: A rounded rectangle labeled "MCP Server" with an icon of stacked server racks/databases. * **Connections between Top Level Components:** Dashed lines with arrowheads point from "MCP Client" to "MCP Server". The label "Transport Layers" is positioned between these components and lines. * **Sub-Components (Bottom Level):** * Below "MCP Client": Two rounded rectangles (represented by icons within) labeled "Notification" (bell icon) and "Sampling" (stamp icon). These are connected to "MCP Client" by a solid line branching into dashed lines pointing downwards. * Below "MCP Server": Three rounded rectangles (represented by icons within) labeled "Tools" (gear and wrench/screwdriver icon), "Resources" (stacked files/folders icon), and "Prompts" (document icon). These are connected to "MCP Server" by a solid line branching into dashed lines pointing downwards. * **Other Elements:** A watermark is present at the bottom right corner, positioned partially under the "Resources" and "Prompts" labels. **Textual Information:** * **Title:** Key Components * **Component Labels:** * MCP Client * MCP Server * Notification * Sampling * Tools * Resources * Prompts * **Connection Label:** Transport Layers * **Watermark/Annotation:** 微信公众号：三只大黄 **Diagram Description:** * **Type:** System Architecture Diagram / Flowchart. * **Main Elements:** * **Components:** * A top-level box labeled "Data Sources" containing icons representing the internet (globe with WWW), databases (stacked cylinders), possibly Slack (four-square icon), and GitHub (cat icon). * A central box labeled "MCP Server" containing an icon representing stacked servers. * A lower-level box labeled "Tools" containing an icon of a gear and wrench. * A lower-level box labeled "Resources" containing an icon of three books/folders. * A lower-level box labeled "Prompts" containing an icon of a document with structured content. * **Connections:** * A dashed arrow pointing from "Data Sources" downwards to "MCP Server", labeled "Data Sources". * A dashed arrow pointing horizontally leftwards from the left side of "MCP Server", labeled "To Client". * Solid vertical lines below "MCP Server" connecting to a point where dashed arrows originate. * Three dashed arrows branching downwards from the point below the solid lines, connecting to "Tools", "Resources", and "Prompts" respectively. * **Labels:** "Data Sources", "MCP Server", "To Client", "Capabilities", "Tools", "Resources", "Prompts". * **Other:** A watermark in the bottom right corner. **Textual Content:** * WWW * Data Sources * To Client * MCP Server * Capabilities * Tools * Resources * Prompts * 公众号: 三只大黄