Evaluate how LLM agents actually use your MCP server — recording proxy, claude/codex, optional LLM judge, HTML report. - View it on GitHub