why is MCP events still in draft? the author literally works at Anthropic.
all I want is:
- system changes (Github, Slack, etc)
- MCP server pushes an event
- agent wakes up and gets to work
@http_samc@rauchg tbh what’s really useful is knowing which part of your app the usage comes from ie. some way to group by labels.
kinda like traces but not as heavy
every AI code review loop:
- reviewer agent flags a bug
- agent fixes it
- reviewer agent finds something new
- repeat until someone gives up
Anthropic's fix: hard cap at 3 rounds, then the reviewer escalates from "fix this" to "why does this exist"