Begin recording a mobile interaction session. Initializes session memory, monitors the UI hierarchy, and starts capturing network events. Returns a session ID to use with subsequent tool calls. During the session, drive the app via execute_ui_action (single steps) or start_flow (stored Maestro yaml). Both update the recording timeline.

Stop the active recording session and synthesize a Maestro YAML test script. Correlates captured UI interactions with network payloads and embeds JavaScript assertions for analytics events.

Capture the current UI element tree from a booted simulator. Works standalone (auto-targets the sole booted device) or within a recording session via sessionId. Returns a normalized accessibility tree with pixel bounds for point-based taps when selectors don't match. Use interactiveOnly to filter to tappable elements.

Dispatch a UI action (tap, type, scroll, etc.) on a target element. Logs the interaction to session memory for later test synthesis. Selector priority: id > accessibilityLabel > text. Scroll/swipe are not supported during a live recording session — use start_flow for complex sequences.

Retrieve intercepted HTTP/HTTPS network transactions for the session from Proxyman. Filter by URL path to isolate SDUI or analytics endpoints. Used to correlate network state with UI state.

Validate that a specific SDUI network response matches expected fields. Returns matched status and a list of any mismatches. Used to assert correct server-driven content is rendered by the UI.

Assert that a set of matching network requests all start within a given time window (e.g., SDUI queries firing in parallel). Fails if fewer than minExpectedCount match or the total span exceeds maxWindowMs.

Assert that a list of expected network calls all fire within withinMs of a referenced UI action. Use to verify that navigating to a screen triggers the right API calls.

Assert that a list of forbidden network calls do NOT fire within withinMs of a referenced UI action. Use to verify cache hits or absence of unnecessary prefetching.

Assert that a set of network calls happened in a specific chronological order. Strict mode fails if any unmatched event appears between ordered matches.

Assert latency budgets: max per-request durationMs and/or max total first-start→last-end across a matcher. Reports p50/p95 stats and excludes events with unknown durations from percentiles.

Assert JSON response fields via dot/bracket paths: equals, contains, exists, type, minLength. More flexible than verify_sdui_payload, which only supports exact field matching.

Assert that requests are not duplicated beyond a threshold. Groups by URL or extracted GraphQL operationName; flags groups exceeding maxDuplicates.

Assert that specific error responses appear in the session. Pair with WireMock stubs to verify the app behaves correctly under injected failures.

Register a recorded session as a named, reusable flow segment. Computes a fingerprint from the correlated steps and saves it to the segment registry for future deduplication.

List available iOS simulators and Android emulators. Filter by platform, state (Booted/Shutdown), or OS version. Use this to discover device UDIDs before calling get_ui_hierarchy.

Get a lightweight mid-session health check showing polling stats, interaction counts, and gap analysis. Use during an active recording to verify the poller is keeping up and interactions are being captured. Only available while session status is "recording".

Discover named Maestro flows in a flows directory (default: ./flows). Each flow is a .yaml file; an optional _manifest.json adds descriptions, tags, and parameter specs. Use start_flow to execute one by name.

Compile an iOS or Android app from source (xcodebuild or gradlew). Returns a taskId immediately; poll via poll_task_status, get final .app/.apk path + bundleId via get_task_result. Default timeout 15 min.

Run a Maestro YAML test file with optional WireMock stub replay. Replays a static script against a booted simulator — does NOT record new network traffic. Returns a taskId; poll_task_status streams live output, get_task_result returns final pass/fail. With MCA_FLOW_PAUSE_RESUME=on, pauses any active recording for the run and auto-resumes; otherwise errors if a session is active. cancel_task interrupts mid-flow.

Execute a named Maestro flow (resolves /.yaml and merges manifest param defaults with caller params). Use to navigate to the area of an incremental change before verifying it. Returns a taskId; poll_task_status streams output, get_task_result returns final pass/fail. With MCA_FLOW_PAUSE_RESUME=on, pauses any active recording for the run and auto-resumes; otherwise errors if a session is active. cancel_task interrupts mid-flow.

Returns current status, duration, and the recent tail of streamed output for a task. Cheap; safe to call frequently. Returns notFound:true for unknown or pruned task IDs (never throws).

Returns the final structured result for a completed task. Returns error/notFound for not-yet-done or unknown tasks. Idempotent — does not consume the task.

Aborts a running task: sends SIGTERM to children, runs registered cleanups in reverse order, marks the task cancelled. Idempotent and never throws.

Lists tasks in the registry, optionally filtered by kind/status. Useful for orphan recovery alongside list_active_sessions.

Install a built app onto a booted simulator/emulator. iOS uses xcrun simctl install; Android uses adb install -r. Returns the resolved bundle id (iOS) when available.

Remove an installed app from a booted simulator/emulator, wiping its storage. Use before install_app to guarantee a clean-state launch. iOS: xcrun simctl uninstall; Android: adb uninstall.

Boot an iOS simulator by UDID and wait for it to be fully ready. Idempotent — returns alreadyBooted=true if already running. Opens Simulator.app by default. Android emulator booting is not yet supported (start it manually).

Capture a PNG of the current simulator/emulator screen; returns an absolute path the agent can read back. Auto-retries on transient failures; returns passed:false on terminal failure instead of throwing.

Run the unit-test target for the project. Returns structured results: pass/fail counts, failing test names, first-line failure messages. Long-running — default timeout 30 minutes.

Execute a declarative feature test in ONE tool call: setup flows → start recording → UI actions → network assertions → stop & compile → teardown. Replaces 8–15 AI-orchestrated tool calls per run with a single deterministic lifecycle. Accepts an inline FeatureTestSpec or a path to a .yaml/.json spec file.

Install a live response-mocking rule via Proxyman. Two modes: staticResponse (return a verbatim payload — feature flags, fixtures) and responseTransform.jsonPatch (proxy to the real backend then mutate the response body in flight — e.g. the loginStatus override pattern). Session-scoped mocks auto-clean on stop_and_compile_test; standalone mocks persist until explicitly cleared. Requires Proxyman running with MCP enabled.

Remove mocks installed by set_mock_response. Pass mockId to remove one; omit to clear all mocks for the session. stop_and_compile_test runs this implicitly on session end.