KestrelError tracking built for AI agents

Single binary, MCP-native, < 50 MB resident. Sentry-style data, designed for LLMs to read instead of dashboards for humans to click.

🤖

AI-first

A built-in MCP server lets Claude Code, Cursor, or your own agent read errors, diagnose them, and mark them resolved with a commit SHA.

📦

Pure Go, embedded SQLite, embedded SPA. < 20 MB stripped. Idle memory < 50 MB. Drop a Docker container or a static binary on a VPS and you're done.

🛡️

SDKs hook the runtime's unhandled-error path and stop there. No fetch patching, no console wrapping, no framework injection. JS SDK is ~1 KB gzip.

✂️

No team RBAC, no alerting channels, no session replay, no APM. The roadmap's non-goals list is firm — feature creep is the failure mode.