Your coding agent tests its own fixes

Your coding agent can now close the loop: read what broke, fix it, run the test again and see the result.

• Start and repeat tests. start_test tests a URL with your instructions, rerun repeats an earlier run, and stop_run stops one. Each test counts toward your plan, with the same limits as the app.
• Wait and compare. wait_for_run returns the verdict when the run finishes. compare_runs lists the findings that are new, fixed or still there.
• More to read. get_steps shows what the browser did and where it errored. get_playwright_test exports a run as a Playwright test. list_projects and get_usage show projects and tests left.
• Your choice per token. New tokens can start tests by default; switch it off for a read-only token. Tokens made before today stay read-only.

Open Connections, MCP to make a token.