The localhost tunnel is the distinguishing feature: check_app_in_browser and probe_page can hit the app you are actively developing, not just a deployed URL, so a regression check runs against the change you just made. The crawl-into-a-knowledge-graph step is what gives later tests context — the agent explores the app once so subsequent evaluations know its pages and flows rather than starting blind each time.
A server for AI-powered browser testing: 8 tools that let an agent browse and verify a live app, probe pages for their rendered state, crawl a site into a knowledge graph, and manage the projects, environments and test suites that runs against.
- check_app_in_browser gives an agent eyes on a running app — it browses, interacts, and reports whether a described task or check passed, on localhost or any URL, for visual QA, flow validation and regression checks
- probe_page returns the rendered state of one or more URLs: a screenshot, page metadata (title, finalUrl, statusCode, loadTimeMs), structured console errors, and a per-URL network summary that collapses refetch loops into one row — the 'did I just break /settings?' smoke test
- trigger_crawl explores pages, UI states and navigation flows systematically and populates the backend knowledge graph, so later evaluations and tests have context about the app
- project manages DebuggAI projects (get, list, create) against a GitHub-linked repo, and environment manages the environments and their login credentials under a project, with passwords never returned
- test_suite lists, creates and runs suites by uuid or name with a project identifier, reporting status and pass rate, and test_case adds and patches individual cases from an agentTaskDescription without auto-running them
- executions is the history — get returns one run in full detail with node executions, state, error info and any screenshot or GIF artifacts, and list pages through past runs by project and status
A DebuggAI API key from debugg.ai, held in DEBUGGAI_API_KEY. Node.js 20.20.0 or later, run as npx @debugg-ai/debugg-ai-mcp over stdio, or from the Docker image. Testing localhost URLs needs the caddy binary — check_app_in_browser, probe_page and trigger_crawl tunnel localhost through a local Caddy proxy, which installs automatically during npm install; if that download did not run, point CADDY_BIN at your own install. Public-URL calls and non-browser tools do not need it.
One command plus a key — npx -y @debugg-ai/debugg-ai-mcp, then supply credentials