Navigate to a staging URL and return factual Semantic JSON of all interactive/test-relevant elements with Playwright-native locators and live state. Use this before writing any Playwright test so selectors are real, not guessed.

Like extract_semantic_dom, but first performs a short DECLARED list of actions (fill/click/press/select/goto/wait) in the main frame, then returns Semantic JSON of the RESULTING state. Use it for post-interaction UI a plain snapshot cannot see: success/error toasts, validation messages, opened dialogs. Derive action locators from a prior extract_semantic_dom call. The page must remain on allowlisted hosts after the actions, or nothing is extracted. Uniqueness reflects capture time — accumulating UI (chat threads, lists) can multiply matches later. The result's observed block lists navigations, xhr/fetch requests (method, path, status), console errors, dialogs and popups seen while the actions ran — use them for waitForURL/waitForResponse.

Diagnostic: navigate to a URL and return its frame tree (frame_path, url, name, same_origin, reachable). Useful for debugging cross-origin iframe boundaries before extraction.

Diagnostic: navigates with the configured QA_MCP_STORAGE_STATE session and reports whether the page bounced to a login-looking path (session likely expired). Use when extractions unexpectedly return login forms instead of the requested page.

The page as a MAP, a few thousand characters: landmark regions (header/nav/main/forms/tables/lists/dialogs) each with a selector you can pass as scope to extract_semantic_dom / session_extract, structured tables (headers, row identity for getByRole('row', { name }), cells) and open dialogs (label/value fields), plus current alert text. Start here on any page you have not seen; then extract one region.

Count every given Playwright expression against the live page (same engine that verified the extraction). Use after writing a test: paste the spec's getBy*/locator expressions and get matches, uniqueness, and the first matched element per expression, plus a summary. Also the drift check to run in CI against a page.

The team's Playwright test-writing conventions as text (same content as the write_playwright_test prompt), for clients that do not surface MCP prompts.

Open a persistent browser session at a staging URL for a MULTI-STEP flow (login → cart → checkout). The page stays open across calls: use session_act to perform declared actions and session_extract to snapshot or diff, then session_close. Fresh context per session (storageState applied if configured). Sessions expire after an idle TTL and are capped in number; the allowlist is re-checked after every step.

Perform a short DECLARED action list (fill/click/press/select/goto/wait, max 20) in an open session's main frame. Returns the page's resulting URL/title and observed: main-frame navigations, xhr/fetch requests (method, path, status — bodies and query strings never captured), console errors, dialogs (auto-dismissed) and popups (recorded, closed). These are the facts for waitForURL/waitForResponse. Derive action locators from a prior extraction (paste its playwright expression). Pass then_extract to get the diff (or a scoped extraction / outline) in the same call. If the actions leave the allowlisted hosts the session is closed and nothing further is extracted.

Snapshot the CURRENT state of an open session as Semantic JSON (same shape as extract_semantic_dom, plus snapshot_id). Pass diff_against: 'previous' (or a snapshot_id) to receive only what changed: added nodes (new toasts/dialogs/fields), removed nodes, changed properties (value, is_disabled, aria_invalid, described_by…) and the behavior observed in between — far smaller than a full re-extraction and exactly the assertion list for the step. Diff identity: frame + test-id, else id, else placeholder, else tag+role+accessible name (+ document-order index); a renamed node with no stable attribute shows as removed + added; primary_locator.playwright changes say which locator is valid in which state.

verify_locators against an open session's current page (after acting, without a fresh navigation).

Close an open session and release its browser context. Always call this when the flow is done. Idempotent: closing an unknown or already-closed id succeeds with was_open: false.

Diagnostic: list open sessions (id, URL, expiry, counts) — recover a session id after losing context.