Eyebrowse

Extract the current page as clean, token-efficient markdown (no LLM involved). Returns the markdown string. If output_path is given, the markdown is written to that file and {path, chars} is returned instead — so you can pick a path and fetch the content from there. You (the agent) do any further structuring/extraction.

Set files on a file <input> by ref (absolute paths). Returns a snapshot.

Fill multiple fields at once. Each field: {ref, value, submit?, clear?}.

Find VISIBLE interactive elements whose text / placeholder / label / type contains `query` — the escape hatch when browser_snapshot is all 'generic' (no usable [ref=eN]). Returns a list of matches, each with a css selector you act on via the ref slot, e.g. browser_click(ref="css=...") or browser_type(ref="css=...", text=...). Example: browser_find("verification code") -> the OTP input + its css ref.

Return a stable CSS selector for an element (by ref) for use in code/tests.

Pre-grant browser permissions so the native prompt never blocks the flow. permissions: list of permission names, e.g. ['geolocation'], ['notifications'], ['camera'], ['microphone'], ['clipboard-read'], ['clipboard-write']. origin: restrict the grant to a specific origin (e.g. 'https://example.com'); omit to apply session-wide. Call this before navigating to a page that requests permissions, or immediately when a permission prompt appears (it will dismiss the prompt automatically).

Accept or dismiss an open JS dialog (alert/confirm/prompt). prompt_text fills a prompt() before accepting. Dialogs stay open until handled.

Finalize and return the HAR file path for a recording session. NOTE: This CLOSES the session — Playwright only flushes the HAR buffer when the browser context closes. The session must have been created with record_har=True. output_path: copy the finished HAR to this path (e.g. 'results/run1.har'). If omitted, the HAR stays at its auto-generated path under data/har/. Non-destructive checkpoint pattern (to keep browsing after capturing traffic): 1. browser_storage_state(path='data/checkpoint.json') # save auth/cookies 2. browser_har_export() # closes…

Draw a magenta outline around an element (by ref) — handy before a screenshot.

Hover over an element by ref. Returns a snapshot.

Type text into the currently focused element — no ref needed. Works on contenteditable rich-text editors (TipTap, Quill, ProseMirror) where browser_type / fill() is a no-op. Also the escape hatch when a coordinate click focused an iframe field and you need to type into it. Pattern: browser_mouse_click(x, y) to focus → browser_keyboard_type(text). delay: ms between keystrokes (simulates human typing speed). Returns a snapshot.

List all open sessions with their current url, tab count, and identity.

Clear all localStorage for the current origin.

Get a localStorage value by key (null if absent).

List all localStorage key/value pairs for the current origin.

Remove a localStorage key.

Set a localStorage key to a value.

Fulfill requests matching a glob pattern with a canned response (response mocking / fault injection). Cleared by browser_unblock_urls.

Click at absolute coordinates (x, y). button: left|right|middle; clicks for multi-click.

Press a mouse button down (at the current position).

Drag from (x1, y1) to (x2, y2) with the left button held.

Move the mouse to absolute page coordinates (x, y).

Release a mouse button (at the current position).

Scroll the mouse wheel by (delta_x, delta_y) pixels.

Navigate to a URL. Returns the page's ARIA snapshot (with [ref=...] handles). wait_until: when to consider navigation done — 'domcontentloaded' (default, fast: DOM parsed) | 'commit' (fastest: response started) | 'load' (waits for ALL resources — slow/unreliable on heavy or proxied sites) | 'networkidle'. Prefer the default and then browser_wait_for a specific element; only use 'load' for simple static pages. timeout_ms: max navigation wait; None = the engine's configured navigation timeout. Auto-creates a session if none exists.

Go back one entry in history. Returns the new page's ARIA snapshot.

Go forward one entry in history. Returns the new page's ARIA snapshot.

Get one request's full detail: request + response headers, and response body (auto-captured for XHR/fetch resource types — body may still be populating for very recent requests). Match by list index or url substring (most recent match).

List network requests seen this session (method, url, status, resource_type). Optionally filter by resource_type (e.g. 'xhr', 'fetch', 'document') or url substring. For full details including headers and body, use browser_network_request.

Start a fresh browser identity in a new session: a novel fingerprint (randomized OS + screen) with isolated storage, optionally paired with a proxy. Proxyless by default; pass proxy_url ('http://user:pass@host:port') or proxy_server[+username/password] to pair an IP. persistent=True mints a reusable profile dir. Returns the new session info.

Create a new stealth browser session and make it current. Returns its id. Runs PROXYLESS by default. To use a proxy, pass either proxy_url ('http://user:pass@host:port') or proxy_server[+username/password]. no_proxy=True forces proxyless even if an env/default proxy is configured. persistent: keep cookies/localStorage in a profile dir across runs. headless: override the configured default (None = use the default). humanize: per-session cursor humanization — a float caps cursor-move time in seconds (e.g. 0.25 = fast but still humanized), True = default speed, False = off, None = use t…

Save the current page as a PDF file. Chromium only. Returns the saved file path.

Press a keyboard key (e.g. 'Enter', 'Escape', 'ArrowDown', 'Tab'). Returns a snapshot.

Reload the current page. Returns the page's ARIA snapshot.

Resize the page viewport (e.g. 1920x1080). Affects layout and screenshot size.

Take a PNG screenshot. By default captures the visible viewport; full_page=True captures the entire scrollable page; ref captures a single element. output_path: if given, the PNG is WRITTEN TO THAT FILE and the absolute path is returned as text (instead of returning the image bytes). Use this to hand a screenshot to an out-of-band vision/OCR tool without routing the (large) image bytes through the caller — e.g. a text-only LLM driver that delegates 'seeing' to a separate vision model.

Scroll the page by pixels in a direction, or scroll an element into view. direction: 'up' | 'down' | 'left' | 'right'. amount: pixels (default 300). ref: if given, scrolls that element into view (direction/amount ignored). Use this to trigger lazy-loaded content or reveal off-screen elements. Returns a snapshot.

Scroll to the very bottom of the page, pausing to let lazy content load. Stops when the page height stops growing (infinite scroll exhausted or real bottom reached). max_scrolls: safety cap. wait_ms: pause per step. Returns a snapshot of the final state.

Pick `value` from a CUSTOM / searchable dropdown or combobox (country & country-code pickers, styled listboxes) where browser_select_option times out. Opens the trigger `ref`, types `value` to filter, commits the highlighted option by keyboard (ArrowDown+Enter), then VERIFIES the committed value and falls back to clicking the option by visible text. Filter by the COUNTRY NAME (e.g. 'Germany'), not a dial code ('49') which matches the wrong row. Returns {committed, actual} + a snapshot — if committed=false the selection did NOT take.

Select option(s) in a NATIVE <select> by ref (match by value OR visible label). Self- verifies and, on a no-match, returns the available option labels. For a CUSTOM / searchable dropdown (styled div, role=combobox/listbox — most country-code pickers) this will not work — use browser_select_combobox instead. Returns the committed option + a snapshot.

Clear all sessionStorage for the current origin.

Get a sessionStorage value by key (null if absent).

List all sessionStorage key/value pairs for the current origin.

Remove a sessionStorage key.

Set a sessionStorage key to a value.

Set extra HTTP headers sent with every request in this session.

Override the geolocation the page reads (note: geoip already aligns geo to the IP).

Set `value` on a FRAMEWORK-CONTROLLED input (React/Vue) via the native value setter + input/change events. Use this when browser_type / fill silently leaves the field EMPTY (controlled components re-derive value from state and ignore plain typing — e.g. some OTP and masked phone inputs). Returns the resulting value + a snapshot.

Toggle the context's network offline/online (to test offline behavior).

Pin a proxy as the default for subsequent sessions/identities. (Proxies are bound at launch, so existing sessions are unaffected.)

Capture the page's ARIA accessibility tree with [ref=...] handles. This is the primary way to 'see' a page: it lists actionable elements (roles + names + refs) without CSS/markup noise. Pass a ref from here to click/type/hover. Frame-hosted elements get prefixed refs like 'f1e36' (iframe 1, element 36). Shadow DOM elements appear with plain eN refs — the locator engine pierces them. enrich=True ALSO appends a DOM map of visible interactive elements, each with a css selector usable in the ref slot (browser_click(ref="css=...")). Use it when the ARIA tree collapses to bare 'generic' nodes with…

Snapshot a specific child frame directly. Use when browser_snapshot returns an iframe node with empty/collapsed children — snapshotting the frame is reliable (works for same-origin AND cross-origin frames). frame_ref: a frame id ('f1'), an element ref inside the frame ('f1e36'), or the <iframe> element's own ref ('e81'). Returns the frame's ARIA tree with refs rewritten to 'fNeM' form so they're directly usable with click/type/etc.

Solve a captcha on the current page and inject the token (no extension). kind: 'turnstile' | 'recaptcha_v2' | 'recaptcha_v3' | 'hcaptcha' | 'funcaptcha'. website_key is auto-detected from the DOM if omitted (for FunCaptcha that's the public key). provider defaults to the configured one (needs that provider's API key in the environment). Returns the solved token. recaptcha_v3: pass page_action (must match the site's grecaptcha.execute action) and optionally min_score (0.1–0.9); the token is injected and grecaptcha.execute is overridden to return it. funcaptcha (Arkose): funcaptcha_subdoma…

Start a Playwright trace (screenshots + DOM snapshots) for later inspection.

Stop tracing and write a trace.zip (open with `playwright show-trace`). Returns the path.

Save cookies + localStorage to a JSON file. Reload it later via browser_new_session(storage_state=path). Returns the file path.

Switch the active page to the most recently opened popup or new tab. Use this after triggering an OAuth / SSO flow (Google Sign-In, GitHub OAuth, etc.) or any action that calls window.open() or opens a target=_blank link. The tool waits up to timeout_ms for the new window to open, then makes it the active page and returns its ARIA snapshot. To go back to the original tab afterward, use browser_tabs(action='select', index=0).

Manage tabs. action: list | new | select | close (index for select/close, url optional for new). Returns the resulting tab list.

Generate a TOTP (time-based one-time password) from a base32 shared secret. No browser or session needed — this is a pure computation tool. Use it when a site uses an authenticator app (Google Authenticator, Authy, etc.) as the second factor. The secret is the base32 string shown during 2FA setup (usually labeled 'secret key' or encoded in the QR code's otpauth:// URI). digits: code length (default 6). interval: time step in seconds (default 30). Returns the current 6-digit (or N-digit) TOTP code as a string.

Type text into a field by ref (top-frame or iframe ref like 'f1e20'). submit=True presses Enter after. Returns a snapshot. Pass the string as `text`. `value` is accepted as an alias for `text` (a common slip, since browser_fill_form fields use `value`) so the call doesn't hard-error.

Enter an OTP/verification `code` into the field at `ref`, handling MULTI-BOX widgets (one <input maxlength=1> per digit that auto-advances) by distributing the digits across the boxes via the React-tracked setter — instead of typing the whole code into box 1 (which lands the digits in the wrong boxes). Works on a single field too. Returns {boxes, value} (read it back) + a snapshot. After this, still SUBMIT (Enter / a Verify button).

Remove all URL routes added via browser_block_urls or browser_mock_url.

Assert an element (by ref) is hidden/absent. Returns {ok, ...}.

Assert an element (by ref) is visible. Returns {ok, ...}.

Assert some visible text appears on the page. Returns {ok, ...}.

Assert an input/element (by ref) has the expected value. Returns {ok, ...}.

Path to this session's native video recording (Chromium only). Only present when record_video=true was passed to browser_new_session. The .webm is finalized when the session closes (browser_close_session); the path is known beforehand.

Wait for a condition then return a fresh snapshot. Pass exactly one condition: text — text to become visible (searched across all frames including iframes). text_gone — text to disappear. selector — CSS/Playwright selector to become visible. url — URL glob/regex to wait for (SPA client-side navigation). network_idle — wait until no network requests for 500ms (SPA render complete). value_ref + value — wait until the element at value_ref HOLDS `value` (substring). Confirms a field/dropdown actually committed (e.g. value_ref=<country picker>, value='Germany'). time …

Wait for a file download to complete and return its saved path. Call this before (or immediately after) clicking a download button. The first download event is captured, saved to save_dir, and the absolute file path is returned. timeout_ms: max wait (default 30s).

Return all WebSocket connections opened this session and their messages. Each entry: {url, closed, messages: [{dir: 'sent'|'received', data: str}]}. url_contains: filter to connections whose URL includes this substring. Useful for real-time apps (trading dashboards, chat, live data feeds) that deliver state over WebSocket rather than HTTP.