This MCP server juggling problem is real - I've definitely been there with the constant re-configuring across different AI apps. It gets old fast. The dynamic tool fetching sounds clever but I'm wondering about performance. If you're looking up tools on-demand instead of having them ready to go, doesn't that add latency to every request? Especially when you're chaining multiple tools together like in those example prompts. Also curious how you handle auth across all these different services. Some tools need OAuth, others use API keys, some have weird custom auth flows. Managing all that centrally seems like it could get messy quick.