Connect to the browser via the extension and report its state

Drop the connection to the browser (the browser itself keeps running)

Navigate a tab to a URL and wait for it to settle

Go back or forward in the active tab's session history

Click an element by selector, visible text, or ref_N from read_page (trusted CDP click with DOM fallback)

Type text into an input/textarea/contenteditable (optionally char-by-char)

Move keyboard focus to an element (needed before press_key on custom widgets)

Send keyboard keys ("Enter", "Control+a"); repeats the sequence times times

Scroll the page or an element (use down repeatedly for infinite scroll)

Get the current page's URL, title, loading status, scroll state, interactivity, and CAPTCHA presence

Extract readable text or raw HTML from the page or an element

Build an accessibility snapshot of the page with stable ref_N ids for click/type targeting

List elements of a kind (links, buttons, inputs...) with names and reusable selectors

Inspect an element's tag, attributes, css path, subtree, and (optionally) HTML

Capture a PNG of the page (full page) or one element; saved under data/screenshots

Save the current page as PDF (A4, backgrounds on); saved under data/screenshots

Run JavaScript in the page and return its result. Use return for a value; secrets in output are redacted unless redact=false. DANGER: requires confirm=true

Register a named persistent script that replays on every navigation

Send JSON data to an injected script and await its reply

Poll the DOM until a CSS selector or visible text appears (or timeout)

Wait for the page to reach a load state

List, open, switch, close, or inspect tabs

List browser windows, focus one, or close one

Check whether the current page shows a CAPTCHA (recaptcha/hcaptcha/turnstile/geetest/funcaptcha)

Wait up to timeout_ms for the user to solve the page's CAPTCHA

Control HTML5 video/audio playback: play, pause, toggle, mute, unmute, seek, set_speed, set_volume, fullscreen, exit_fullscreen, get_info

Run a web search in the active tab and return the top results

Search across all open tabs by title or URL (TF-IDF ranked)

Start capturing HTTP traffic on the active tab via CDP

Stop network capture and return everything captured so far

Peek at captured requests without stopping capture

Send an HTTP request through the browser profile (cookies/session apply)

Get cookies (filter by domain/name), set cookies, delete/clear (auto-backup first), export/import encrypted snapshots

Add a bookmark (updates title/tags if the URL already exists)

Delete a bookmark by id (or by exact URL)

Search bookmarks by keyword and/or tags

List all saved bookmarks

Search recorded navigation history (most recent first)

Hover over an element to reveal hover menus, tooltips, or submenus

Unified interaction: click, double_click, right_click, move, type, fill, key, scroll, hover, wait, navigate, screenshot

Server status, connection transport, store sizes, live refs, and uptime