# OpenAI给MCP补上云托管：写完就能做成ChatGPT插件 OpenAI 正在把 MCP（Model Context Protocol）从"本地跑一跑"推向"云端一键部署"。 ## 核心变化 过去，开发者想让自己的 MCP Server 接入 ChatGPT，通常需要： - 自己找服务器托管 - 配置公网地址 - 处理鉴权、证书、扩缩容 - 手动在 ChatGPT 里注册连接器 现在 OpenAI 补上了**云托管**这一环：开发者写完 MCP Server，可以直接托管到 OpenAI 侧，然后作为 ChatGPT 插件（Connector）对外提供。 ## 这意味着什么 **1. 门槛大幅降低** 从"写代码 + 运维 + 上架"变成"写代码 + 点部署"。个人开发者和小团队也能快速把自己的工具变成 ChatGPT 可调用的能力。 **2. MCP 生态加速** MCP 原本是 Anthropic 主推的开放协议，OpenAI 这次不仅支持，还直接提供托管基础设施。等于用行动承认了 MCP 作为"AI 工具调用标准"的地位。 **3. ChatGPT

Beating AI News: OpenAI adds MCP Server hosting to ChatGPT Sites. Previously, if you wrote an MCP yourself and wanted cloud-based ChatGPT to call it, you usually had to find a server to deploy it on. Now you can run the MCP directly on Sites, turn it into a ChatGPT plugin, and after installation it works on web, mobile, and desktop.

OpenAI's Max Stoiber demonstrated completing the entire process with a single sentence. He asked Sites to "make a to-do list that can be used in ChatGPT," and the system created an MCP Server, deployed it to Sites, then generated a plugin and installed it into ChatGPT.

This mainly solves the deployment problem of local MCPs. If an MCP only runs on your own computer, then once the computer is turned off, cloud-based ChatGPT cannot access it. After switching to hosting by Sites, the MCP can keep running in the cloud, and developers no longer need to maintain the server themselves.