Skip to content

Audit tools

Six tools for cloud audits — accessibility runs driven against saved journeys on a dedicated VM with a real screen reader.

One identifier runs through all of them: the run id. start_agentic_audit returns it the moment the audit is asked for, and get_audit_progress, get_agentic_audit_report and cancel_audit all take it.

Start an audit on the cloud runner.

NameTypeRequiredNotes
projectIdstringconditionalRequired for a project-wide audit without journey or target ids.
journeyIdsstring[]noSaved journeys to test. Omit or leave empty to run all.

Returns: a text summary naming the run id of every journey the audit covers (and the VM name when one is reported), plus the instruction to poll get_audit_progress. Each run is queued from the start; the VM takes roughly 4–5 minutes to boot before it is running. When every audit slot on the plan is busy, the summary says so and names the run’s place in line.

Returns an error block reading Failed to start audit: no run returned if the audit was not accepted.

A journey with a persona signs in as it. When a persona has no login captured on the run’s deployment, the audit 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.

Get the live status of an audit by its run id.

NameTypeRequiredNotes
organization_idstringyes
run_idstringyesFrom start_agentic_audit.

Returns: the run — its id, a status, a reason when it ended without finishing, queuePosition, and violations: how many tickets the run found, each counted once however often the run found it. queuePosition is the run’s place in line while it waits for a slot, and null once it holds one. An unknown run id is an error block. There are six statuses:

StatusMeaning
queuedWaiting for a slot, or its VM is booting. Keep polling.
runningIn progress. Keep polling.
completedFinished. A report is available.
completed_with_errorsFinished with partial coverage. A report is available.
failedStopped without producing a report.
interruptedStopped — including by cancel_audit — without producing a report.

Poll while the status is queued or running. Treat the other four as final: waiting for completed alone leaves you polling forever on a run that ended as completed_with_errors. A run whose VM never started ends failed with the reason never_started.

Use cancel_audit in the generated tool reference.

See the generated operation reference for arguments. Pass runId; the server derives its organization. The response is GraphQL JSON under agenticAuditReport, or null when no run has that id. A run still going answers status: running; a run past its 30-day retention answers its findings and tickets without results or agentNotes.

The owner returns deduplicated tickets, findings, agent results, and agentNotes as Relay connections. Each has edges and pageInfo. While a connection’s hasNextPage is true, pass its endCursor as the corresponding ticketsAfter, findingsAfter, resultsAfter, or notesAfter argument. first defaults to the repository’s 100-item page size.

Ticket screenshotUrl comes from the API and is null when there is no screenshot key. A URL does not prove the stored image exists. Use get_violation_screenshot to retrieve it.

recording is returned in the JSON. A stored recording includes its signed url, contentType, sizeBytes, and expiresAt. An uploading recording names retryAfterMs. No recording is null.

Get a screenshot image and its link for an accessibility violation.

NameTypeRequiredNotes
organizationIdstringyesRaw or GraphQL organization global ID.
keystringyesA ticket’s screenshotKey, from an audit report or list_agentic_tickets.

Returns: an image content block containing the original base64 data and stored MIME type, followed by a resource_link content block. The link’s uri is the Binclusive API’s screenshot URL, its name is the screenshot key, and its mimeType matches the image. JPEG and PNG retain their stored MIME types.

Returns GraphQL JSON { "agenticViolationScreenshot": null } when retrieval finds no image, including when the key belongs to another organization. That result contains no image or link.

List a project’s run history.

NameTypeRequiredNotes
organization_idstringyes
project_idstringyes
firstnumbernoPage size. Default 20.
afterstringnoCursor from a previous call’s pageInfo.endCursor.

Returns: a paginated connection merging two entry kinds, each naming itself in __typename:

__typenameWhat it is
AgenticRunA cloud audit Binclusive drove.
CliScanA static scan your CI reported via b8e ci.

To read the findings a b8e run uploaded, pass its CliScan id as scan_run_id to list_scan_run_findings.

Entries are ordered newest first, across both kinds. History grows a row per CI run, so ask for the page you need and walk back with after only if the answer is not in it.

Runs that stopped without producing a report — failed and interrupted — are not listed here. That includes a run you stopped with cancel_audit. To check on one, use get_audit_progress with its run id.