A viral HN post this week called the MCP explosion a "distributed systems nightmare." The argument: we now have hundreds of MCP servers, agents call them freely, and nobody has solved the coordination problem of which tool to call when. The result is infinite loops, burned credits, and solo founders staring at a $20 hole in their API balance after lunch.
I know this problem personally because I caused it.
The $20 Lunch Break
I was building an educational project last year. The code had a loop that retried API calls when the response came back null — standard defensive pattern. Except the responses were not actually failing. They were not being saved properly. The storage code was broken, so the loop kept seeing null, kept retrying, kept calling Anthropic.
I went for lunch. Came back an hour later. Entire $20 in credits gone.
