I made an attempt at making an agent first mcp server manager. I was inspired how cline does this, I found the UX of asking an agent to maintain its own mcp servers to be nice. But I didn’t like that it almost always ends in keys being exposed to the agent conversation (even if there are other safeties in place). This solution tries to maintain that the agent first UX while making efforts at security and ease of use. Feedback and contributions please.