Skip to content

Run Agentic QA journeys in the cloud

A cloud QA run performs a project’s Agentic QA journeys in a real browser on a dedicated virtual machine and judges each step against its expected result. It is asynchronous: you start it, poll it, and read the verdict when it ends. Every step below is an MCP tool call.

You need a project id, from list_projects.

QA journeys are not the accessibility journeys a cloud audit runs, and their ids start with qaj_. List the ones the project has:

list_qa_journeys(project_id)

To read one journey’s steps before you run it:

get_qa_journey(qa_journey_id: "qaj_…")

If the project has none, import them. Each journey has a name, a start page and its steps:

import_qa_journeys(
project_id,
journeys: [
{
name: "Sign in",
startPath: "/login",
steps: [
{ instruction: "Type the demo email and password and submit", expectedResult: "The dashboard opens" }
]
}
]
)

The answer lists each created journey’s qaj_… id. The import is all or nothing: a refusal names its code and creates nothing. Importing a journey whose sourceKey matches an existing one updates it.

start_cloud_qa_audit(journeyIds: ["qaj_…", "qaj_…"])

Give 1 to 5 journeys; they run one after another on one VM. A web project runs them in a browser. An Android project runs them on an Android emulator, and needs the uploaded build to install as build_target:

start_cloud_qa_audit(journeyIds: ["qaj_…"], build_target: "1.4.0/a1b2c3")

Keep the payload’s journeyAudit.id to follow the audit. Keep each cloudRunRequestId to cancel its VM request. A group also returns one cloud audit id for all of its members.

The call starts nothing and returns an error block when:

  • the project does not have the qa-audits capability. For a web project with Agentic QA turned off, the block reads “Web QA audits aren’t available yet — the qa-audit-enabled flag is off for this organization.” This is the same sentence the dashboard and CLI show.
  • the organization’s cloud QA runs are already in use. Wait for one to finish and start again.
  • the project has no single website to start on. Set the project’s site and start again.
  • an Android project has no build_target, one that is not <release>/<dist>, or one with no uploaded app. Upload the build and name it. A web project refuses build_target.
get_journey_audit(id: journeyAudit.id)

Wait for the owner’s pollAfterMs between reads until done is true. Each QA member says whether it is waiting, completed, failed or did not start. Finished runs include their verdict, steps and findings. Failures include the owner’s message. The owner’s qaVerdict gives the batch result, including members that never started.

cancel_cloud_qa_audit(cloud_run_request_id)

The answer says whether it was cancelled, had already ended, or was not found. A run that had already finished is left as it was.

List the project’s QA runs, newest first, each with its verdict:

list_qa_runs(project_id)

Then read one run step by step:

get_qa_run_result(runId: "qar_…")

Read the run’s verdict and its steps’ verdicts:

  • findings: the run raised a QA Finding. That is a product defect to report.
  • clean: every step was checked and passed.
  • incomplete: the run is still going, or a step went unchecked. While it runs, read it again later.
  • failed or site_blocked: the run stopped before a result, or the site’s bot protection blocked it. Report a finished run that is not clean or findings as unchecked, not as a defect.
  • A journey you asked for with no run in the progress list never started.

The result’s findingIds are the QA Findings the run raised (qaf_…). Read each one:

get_qa_finding(qa_finding_id: "qaf_…")

It names the step’s instruction, its expectedResult, what the page observed instead, the reasoning behind the failure, the page url and a screenshotUrl. To see every QA Finding in the project across runs:

list_qa_findings(project_id)

The runs and findings are also stored on the project, under Agentic QA in the dashboard and in b8e qa audit list.

For each tool’s full arguments and fields, see the cloud QA, QA journey and QA Finding tools reference. For what each step outcome means, see About Agentic QA.