Cloud QA tools
Tools for Agentic QA runs — QA journeys driven in a real browser. Three start, cancel and follow a cloud run on a dedicated VM; two read a project’s QA runs and one run’s result, wherever it ran.
A start returns a cloud audit id to follow with get_journey_audit. Keep each cloud run request id
for cancel_cloud_qa_audit. A QA run’s own id (qar_…) appears when its VM starts the journey;
list_qa_runs and get_qa_run_result work on that id.
start_cloud_qa_audit
Section titled “start_cloud_qa_audit”Start QA journeys on the cloud runner. A web project’s journeys run in a real browser on a browser
VM; an Android project’s run on an Android emulator VM, which installs the build build_target
names first.
| Name | Type | Required | Notes |
|---|---|---|---|
qa_journey_ids | string[] | yes | 1 to 5 QA journeys of the project. They run one after another on one VM. |
build_target | string | Android only | The uploaded build as <release>/<dist> (e.g. 1.4.0/a1b2c3). Required on an Android project, refused on a web project. |
Returns: a text summary naming the cloud run request id and the VM, plus the instruction to
poll get_journey_audit and the cancel_cloud_qa_audit call that stops the run. The VM takes
a few minutes to boot before the first journey starts. When every concurrent-run slot on the plan is
busy, which accessibility and QA cloud runs share, the summary says the run is queued and names its
place in line instead of a VM; it starts on its own when a slot frees.
Returns an error block, without starting anything, when the project does not have Agentic QA turned
on: the tool reads the project’s cloud capabilities first and, when qa-audits is not among them,
answers with the sentence the dashboard and CLI show for that capability. An unknown project or a
project of an unknown platform is refused the same way. It also returns an error block with
billing’s own sentence when billing refuses the run for a reason other than a busy slot, or when the
project has no single website to start on. On an Android project it returns an error block naming
build_target, without starting anything, when build_target is missing, is not
<release>/<dist>, or names a build with no uploaded app.
A QA journey with a persona signs in as it. When that persona has no login captured on the run’s
deployment, the run is refused before it starts, with an error naming the journey and the persona.
The fix is b8e profile create --persona … --target … on the person’s own machine; MCP captures
no login. list_personas shows which deployments each persona has a
login on.
cancel_cloud_qa_audit
Section titled “cancel_cloud_qa_audit”Cancel a cloud QA run by its request id. The request id carries its project, so any member of that project can cancel it.
| Name | Type | Required | Notes |
|---|---|---|---|
cloud_run_request_id | string | yes | From start_cloud_qa_audit, as it was returned. |
Returns: one of three answers:
- Cancelled: the cancellation was accepted.
- Already ended: the run had finished before the call, so nothing changed.
- Not found: an error block naming the id, when no cloud QA run has that id.
get_journey_audit
Section titled “get_journey_audit”Read the cloud audit returned by a single or group QA start.
| Name | Type | Required | Notes |
|---|---|---|---|
id | ID | yes | The start payload’s journeyAudit.id. |
The answer contains frozen members, done, pollAfterMs and qaVerdict. Poll after
pollAfterMs until done. QA members answer waiting, completed, failed or not started.
A completed or failed member carries its run and the owner’s verdict; a failure or a member
that did not start includes the owner’s message. qaVerdict is the batch verdict when done.
The caller does not match run membership or combine run verdicts.
list_qa_runs
Section titled “list_qa_runs”List a project’s QA runs, newest first.
| Name | Type | Required | Notes |
|---|---|---|---|
project_id | string | yes | |
first | number | no | How many runs to return, 1 to 100. Default 20. |
after | string | no | pageInfo.endCursor from the previous call, to load the next page. |
Returns: runs, totalCount and pageInfo (hasNextPage, endCursor). Each run has its id
(qar_…), journeyId, name, state (running, completed, failed or abandoned), verdict
(clean, findings, failed, incomplete or site_blocked) and startedAt. A project with no
runs returns an empty runs. An unknown project is an error block.
get_qa_run_result
Section titled “get_qa_run_result”See the generated operation reference for arguments. Pass runId,
a qar_… ID. The response is GraphQL JSON under qaAuditRun, or null for an unknown run.
The owner returns the run’s state, verdict, start location, and joinedSteps in journey order.
Each joined step has its instruction, current state, verdict, and kind. A recorded step’s
recorded field carries execution, evidence, observed text, reason, and screenshot URL. A pending
step has recorded: null, including a step a closed run never reached. findings contains the QA
Finding IDs. Per-step tester actions are not included.