Your agent writes the code.
Test Maze grades it.
Connect Claude Code, Cursor or any MCP coding agent. Keep the test plan, record every run against the exact commit, and get a pass/fail verdict computed by fixed rules.
$ npx -y @testmaze/mcp init tmt_…
✓ feature.verify — 12 test casescheckout
✓ testrun.create — a4f9e2cfeat/checkout
✓ pdlc.verify — passship
12 passed · 0 failedverified by Test Maze