Notes from building M3
What we learn testing MCP servers against real coding agents, with the runs to back it up.
Claude Code and Codex break on different MCP features
We ran Claude Code and Codex against the same ten MCP servers at pinned versions. Five features split them, and they never failed the same one.
Your MCP server passed. Did the agent call it?
Test an MCP server with plain pytest, break it on purpose, then run the same tests through real Claude Code and Codex. A walkthrough with M3.