Move debugging into the system: cheaper monitoring, less ops time

Hunting a memory leak turned into a pattern: push the debugging logic into the process, accumulate context in RAM, emit only on error. The payoff is on two axes — less operator time chasing incidents, less monitoring spend on the routine path.

May 24, 2026 · 9 min

Writing log-mcp: an MCP server for log analysis, built with Claude

How I built an MCP server that lets Claude analyze large log files — from first prototype to a Rust classifier doing 1.3M lines/sec — with Claude writing most of the code.

March 1, 2026 · 21 min