The error-cluster tool is the one that changes how a failed run reads: instead of a list to work through, you get the distinct root causes and their sizes. Two other choices are worth noting. The audit flow deliberately keeps your source local — the server sends a prompt and branch signals, your agent reads the code, and only the finished report goes back — and connect_integration returns an OAuth URL with an instruction to show it to the user rather than open it, which is the right shape for an authorization an agent should not complete on its own.
A client for TestDino covering 35 tools across nine domains: automated run results and their AI insights, manual test cases and suites, releases, manual runs, exploratory sessions, failure triage and issue-tracker integrations. One personal access token reaches every organization and project you have permission for.
- Test runs browsed by branch, status, author, commit, environment and time interval, with per-run statistics, suites, cases, git metadata and error detail — up to 20 runs fetched in one batch (list_testruns, get_run_details).
- Test cases listed either inside one run or across runs by branch, author, commit or environment without looking the runs up first, and a single case's error message, code snippet and artefacts (list_testcase, get_testcase_details).
- Failure triage at scale: a run's failing tests grouped by error signature, so ten reds sharing one root cause read as one problem rather than ten (get_run_error_clusters).
- Debugging history for a case — every past execution with the run, counter and branch it belongs to — and TestDino's AI insights at three levels, from a run's failure categorisation down to concrete fixes with code snippets (debug_testcase, get_ai_insights).
- A trace-driven debugging runbook for a failing Playwright test, teaching the exact trace CLI protocol from open through actions and snapshot to close (get_trace_analysis).
- A code-quality audit loop that never uploads your source: fetch the curated audit prompt with branch signals for top failing, flaky and slow tests, analyse locally, then submit the finished report with a score (get_audit_report, submit_audit_report).
- Manual test cases created, updated, searched and read in full — steps, preconditions, custom fields, comments, the latest 20 versions and the latest 100 execution results — organised into a nestable suite hierarchy (list_manual_test_cases, get_manual_test_case, create_manual_test_case, update_manual_test_case, list_manual_test_suites, create_manual_test_suite).
- Releases that nest up to three levels deep, with dates, linked issues and rolled-up run and status counts, addressable either by internal ID or by a counter like MS-12 (list_releases, get_release, create_release, update_release).
- Manual runs scoped to the whole project or to selected cases and suites, attached to a release, with per-case assignment by email or user ID and verdicts set to passed, failed, blocked, skipped or retest (list_manual_runs, get_manual_run, create_manual_run, update_manual_run, list_run_test_cases, update_run_test_case).
- Exploratory sessions with a mission written as rich HTML, an assignee, an estimate in minutes and a linked release (list_sessions, get_session, create_session, update_session).
- Issue-tracker integration end to end: check whether a provider is connected, get an OAuth URL to show the user, file an issue from a specific TestDino run or case, and check later whether it is still open (get_integration_status, connect_integration, create_external_issue, get_external_issue).
- A connection check that confirms the token, names the account, and lists the organizations and projects it reaches along with your role in each (health).
A TestDino personal access token from User Settings, Personal Access Tokens, passed as TESTDINO_PAT; launch with `npx -y @testdino/mcp` over stdio. Your role decides what works — the write tools need org_owner, org_admin or org_member, so a viewer gets the reads and nothing else. Trace analysis additionally needs the Playwright agent CLI at 1.59 or newer on your machine.
One command plus a key — npx @testdino/mcp, then supply credentials