Bugbrain

Fetch a single project by id (name, websiteUrl, environment, whether credentials are configured).

Diff a run’s issues against the previous run for the project: new / resolved / still-present.

Fetch the LLM-as-judge trust report for a run: per-issue validity (real / false-positive / insufficient-evidence), failure root-cause, and an overall run-quality score (1–5). Set generate=true to (re)generate it (spends LLM budget). The honest answer to "can I trust this run?".

Fetch a run’s status, counters, business context, and discovered test plan.

Fetch the step-by-step agent trajectory (thought / action / observation) for a run.

Fetch a test-plan run rollup: items total/passed/failed/skipped, status, duration.

Fetch the consolidated, severity-categorized report for a Web-Test run: status, the swarm config (breadth / agent count), the merged bug report, and the issue count. Poll until status is COMPLETED.

List AI test cases for a project. Optionally filter by kind (e.g. PROMPT_CONSISTENCY, INJECTION_PROBE, HALLUCINATION_SCAN) or status.

List recent AI test runs for a project, optionally filtered by test case. Returns status, key metrics (consistency score, hallucination risk), and timestamps.

List the organization’s API keys (id, name, scopes, created/last-used). Never returns the secret.

List the environments available for a project and which one it currently targets (production / staging / development).

List all LLM/AI test targets configured for a project. Each target represents an AI system under test — a chatbot, voice agent, or LLM endpoint.

List issues for a project, with severity, category, status, page URL, and AI confidence.

List the projects (web apps under test) in the connected organization.

List the issues a run found, with severity, category, page URL, AI confidence, and screenshot links.

List recent runs for a project (or across the org when projectId is omitted).

List curated test cases for a project (or across the org when projectId is omitted).

List test plans (collections of test cases) for a project, or across the org.

List the organization’s outbound webhook subscriptions (url, events, enabled).

Create a new least-privilege API key. Pass `scopes` to restrict it (empty = full access). The secret is returned ONCE — store it immediately.

File a compliance finding (by its signature) into the org’s connected issue tracker (Jira or Linear). Returns the created issue key + URL.

Subscribe a public https URL to run events (run.completed / run.failed) for push-driven CI. Returns the signing secret ONCE — store it to verify the HMAC signature on delivery.

Permanently revoke an API key by id.

Run a dedicated WCAG 2.1/2.2-AA accessibility audit (axe-core + agent verification). Returns immediately with a run id — poll get_run_status, then get_accessibility_report for the score.

Trigger execution of an AI test case. Returns immediately with a run id — poll get_ai_test_results for the outcome. Requires the org to have an available AI test quota.

Run a compliance scan (SOC2 / GDPR / ADA / WCAG 2.2 AA / ISO 27001 / EU AI Act) that maps findings to framework controls with evidence. Returns a run id immediately — poll get_run_status, then get_compliance_report.

Run a TWO-USER test: two agents drive the app at the same time as different users to surface permission leaks (one user seeing another's data), broken collaboration, and state races. actorA logs in with the project's configured credential; provide actorB's login inline (encrypted at rest). Describe the collaboration scenario to exercise. Returns a runId; poll get_web_qa_report. Metered as one Web-Test run. Provide EXACTLY ONE of `url` or `projectId`.

Point Web-Test at a URL or an existing project and a swarm of QA agents explores and bug-tests the site — clicking flows, filling forms, probing edge cases — and files what it finds as issues. Returns a runId immediately; poll get_web_qa_report for the consolidated, severity-categorized bug report. For live progress use get_run_status / list_run_issues / watch_run (a Web-Test run is a normal run). Provide EXACTLY ONE of `url` or `projectId`. For a localhost / not-yet-deployed site, create a public tunnel first with create_tunnel and pass the tunnel URL.

Switch the environment a project targets (production / staging / development).

Kick off an AUTONOMOUS exploratory test run: the AI agent explores the site, discovers flows, and finds bugs/accessibility issues nobody scripted. Returns immediately with a run id — poll get_run_status (or use watch_run) for progress.

Start a run intended for regression: explores like an exploratory run, then use get_run_comparison to diff issues against the previous run (new / resolved / still-present).

Given the files changed in your diff, pick the most relevant APPROVED test cases (by token overlap vs title/module/tags) and run them. Returns the selection + the queued executions. Set dryRun=true to only see the selection. Poll get_execution_result for each verdict.

Fetch one issue’s full detail incl. the AI triage: root-cause hypothesis, impact area, reproduction confidence, an engineering summary, and a suggested fix.

Stream a run’s live progress as MCP progress notifications until it finishes, then return the final run state. Use for interactive sessions; prefer start + get_run_status polling for fire-and-forget automation.