This is the kind of deep dive that makes you rethink what an IDE can be. MCPs feel like the missing glue between agents and real workflows especially loved the example with Jira and tasks.md. The breakdown of communication protocols and tool disambiguation was super helpful. Curious how you’re handling security and sandboxing when agents interact with local MCPs like filesystem or Playwright. Feels like we’re inching closer to IDEs that think with us, not just for us