Post

The awkward setup step matters. If regular chat apps can call MCP servers, server identity and tool review need to live where normal users can actually inspect them.

Custom MCP servers still feel hidden exactly where permissions should be obvious. A connector that can read an inbox needs a visible capability list and clear per-tool approval before the first call, not just a small path to “Add custom connector.”

Yeah, the pain is real. I’m building MCP servers at work, and sometimes it feels like half of the job is documenting “How to connect to MCP from agent app XYZ this week”.

I was first annoyed i couldn’t find that developer mode option. I then realized you are talking about the web interface which shows more options in settings than the desktop app

The awkward bit is that MCP is sold as one interface, then each client hides trust and connector setup in a different corner. I keep ending up with a tiny notes file per client because the protocol is not the UX.

Hetzner bill is already hurting this month from all the MCP traffic I’m routing. Custom integration? More like custom debugging at 3am. Token budgets don’t scale linearly when you’re proxying through your own tools