📄 中文摘要
构建了一个基于 MCP 的受管控 AI 代理系统，该系统通过工具执行现实世界的操作，但始终在严格的政策控制下进行。该系统不仅关注代理可以做什么，更强调代理被允许做什么以及必须阻止的操作。核心思想是将 MCP 作为能力层，将 Actra 作为治理层，MCP 提供真实工具（如 Notion 工作区操作），AI 代理选择并调用这些工具，而 Actra 在执行之前评估每一个工具调用。
我尝试了 Notion MCP 挑战——我能控制我的 AI 代理吗？
出处: I Tried the Notion MCP Challenge — Can I Control My AI Agent?” ⭐
发布: 2026年3月29日
📄 English Summary
I Tried the Notion MCP Challenge — Can I Control My AI Agent?” ⭐
A Governed MCP-Based AI Agent System has been developed, where real-world actions are executed through tools under strict policy control. The focus is not only on what agents can do, but also on what they are allowed to do and what must be blocked. The core idea involves using MCP as the capability layer and Actra as the governance layer. MCP exposes real tools, such as Notion workspace actions, which the AI agent selects and invokes, while Actra evaluates every tool call before execution.