模型上下文协议 (MCP) 是一个开放标准，用于定义应用程序与大型语言模型 (LLM) 共享上下文的方式。 MCP 提供了一种将 AI 模型与不同的数据源和工具连接起来的标准化方式，使它们能够更高效地协同工作。
You can manage MCP server usage in your organization or enterprise by configuring MCP policies on GitHub.
The MCP servers in Copilot policy defines whether MCP servers can run at all across Copilot clients. We recommend keeping this policy enabled and, if necessary, restricting the MCP servers that users can run to an approved list.
MCP allowlists
The recommended method for creating an allowlist is to use your enterprise's managed-settings.json file. This allows you to apply settings across clients that users cannot override.
Alternatively, you can host your own MCP registry and restrict access to servers in the registry. However, this method has weaker enforcement than managed-settings.json.
| Method | Managed settings file | Custom registry |
|---|---|---|
| Release phase | Generally available | 公共预览, not prioritized for development |
| Ease of setup | You can host a configuration file on GitHub that applies automatically to clients. | You must host your own registry that matches the MCP specification and serves HTTPS requests. |
| Enforcement level | Enterprise-wide settings, overridable for enterprise teams | Enterprise-wide or for individual organizations |
| Supported clients | Clients supported by the managed-settings.json file, see Getting started with enterprise-managed settings. Planned to expand in the near future. | See MCP 专用注册表强制实施. |
| Server matching method | Secure matching based on name, URL, or stdio commands | Less secure matching, based on name or ID only. Users can bypass the restriction by editing configuration files. |
Next steps
To configure an allowlist on GitHub, see 为企业配置 MCP 服务器允许列表.