Cloudroots

• Inspect website — Fetch public HTML and return observed controls and constraints. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_test_case, review_test_cases, generate_report, export_report. Phase 1 is read-only and does not click, submit, pay, or delete. Does not execute forms. Private networks and non-http(s) URLs are blocked.
• Generate test cases — Turn observations into functional, negative, and boundary cases with concrete synthetic data. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_test_case, review_test_cases, generate_report, export_report. Phase 1 is read-only and does not click, submit, pay, or delete. Empty input returns a message, not a silent empty list.
• Generate boundary cases — Build concrete boundary values from one observation's min, max, minLength, or maxLength. If no limit was observed, returns EMPTY instead of a generic sentence.
• Validate test case — Check structure, evidence, honesty of basis, and destructive-action approval. Returns PASS, FAIL, or NEEDS_REVIEW. Malformed cases return INVALID_TEST_CASE instead of crashing.
• Detect duplicates — Find exact and semantic duplicates. Cases for different targets are never duplicates, even when the wording is similar.
• Analyze observation coverage — Report observation coverage for testable controls only. This is not requirements coverage. Page and form observations are excluded from the percentage.
• Score test case — Score completeness, expected-result text, evidence honesty, and safety on a 1-5 rubric. MIN_QUALITY_SCORE (default 3.5) sets meetsThreshold.
• Queue test cases for review — Mark cases PENDING, or NEEDS_EDIT when they score below the threshold. This tool cannot approve or reject.
• Generate functional testing report — Summarize observations, validation, quality, duplicates, coverage, and destructive cases. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_test_case, review_test_cases, generate_report, export_report. Phase 1 is read-only and does not click, submit, pay, or delete.
• Compare two inspections — Compare observations from an earlier inspect_website run with a newer one. Lists controls that were added, removed, or changed (required, limits, options, labels) and the existing test cases that need re-running. Use it for regression checks after a release.
• Check links — Request each unique link on a public page with a real GET and report the HTTP status, redirects and timing. Returns verified runtime results (basis verified_runtime_result), not inferences. Same-origin links are checked first. Read-only, nothing is submitted. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_test_case, review_test_cases, generate_report,
• Audit page quality — Fetch a public page and check title, meta description, lang, viewport, h1, canonical, image alt text, form labels, link text, duplicate ids, HTTPS and security headers. Returns findings plus observed_fact test cases. Read-only. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_test_case, review_test_cases, generate_report, export_report. Phase 1 is read-
• Export test cases to Excel or Word — Build a downloadable .xlsx (Summary, Test Cases, Observations sheets), .docx, .csv, Playwright .spec.ts, or Gherkin .feature file from observations and test cases. Over HTTP it returns a downloadUrl valid for 30 minutes; over stdio it saves the file and returns savedPath. Give the user the link or path. Order: inspect_website, then generate_test_cases, then optionally check_links and audit_page for verified and observed cases, then validate_test_case, detect_duplicates, analyze_coverage, score_t
• Generate multi-step journeys — Build end-to-end journeys from observations: walk the main navigation, follow each call to action, complete a form with valid data, refuse a bad email, refuse an empty form, open and close a dialog. Each journey is a list of steps with expectations. Nothing runs yet; pass the journeys to run_journeys.
• Start proving you own a site — Browser tests that click and type may only run on sites the account owns. This returns a token to publish as a file at /.well-known/cloudroots-verification.txt or as a DNS TXT record. Then call check_domain_verification.
• Check domain ownership — Looks for the verification file or DNS TXT record. When it matches, the host (and its subdomains) can be driven by browser tests for this account.
• List verified domains — Domains this account has started or finished verifying, plus the hosts the operator allowlisted.
• Remove a domain — Stops browser tests from driving this host for the account.
• Save a login for test runs — Save a test account so browser runs can sign in first. The username and password are asked from the person through the client, never through the AI, and stored encrypted. Only works when the client supports private prompts, and only for a host this account may drive.
• List saved logins — Names and hosts only. Credentials are never returned.
• Delete a saved login — Removes the stored credentials.
• Save a run to history — Keep the observations and test cases from a run so it can be compared later with compare_runs. Needs an account.
• List saved runs — Newest first. Includes runs saved by scheduled monitors.
• Get a saved run — Returns the stored observations and test cases for a run id.
• Delete a saved run — Removes a run from history.

Showing the first 25 tools — full list and schemas: Directory API.