Capture what is visually on the surface of the page right now — including modals, overlays, dropdowns, iframes and floating UI. Returns a compact JPEG screenshot (or a text summary of open modals/overlays if textOnly=true for non-vision models). Use this FIRST before snapshot/DOM when you need to understand the current visual state. Much cheaper than a full accessibility tree for orientation.

Control the visible glowing cursor and execute clicks or hovers. This is the single, primary tool for in-page navigation and clicking. Target an element by ref (preferred), selector, visible text, or x/y coordinates without needing manual coordinates — the cursor will automatically move to the element and click it.

Check whether the Chrome extension is connected, list open tabs and show the active tab. Call this first when browser tools seem unavailable.

List, create, activate, close, reload, duplicate, pin or mute Chrome tabs. Actions: list, new, activate, close, get, reload, duplicate, pin, mute, unmute.

Open a URL, go back/forward, reload or stop the active tab. Waits for the requested load state. Use newTab=true to open in a new tab.

Return a compact accessibility-style tree with stable refs (e.g. f0_e12) for interactive elements. Do this before clicking or typing. Refs belong to a frame; use allFrames=true to include iframes.

Capture the viewport, the full page, or one element. Set annotate=true to draw element refs as labeled boxes and grid=true for a 100px coordinate grid. Returns an image the model can see.

Type into an input, textarea or contenteditable element. When a target is provided (ref or selector), the cursor automatically moves to the element and clicks it to activate/focus it first, then executes natural typing via insertText without DOM manipulation. Clears the field by default and uses trusted CDP keystrokes. Set pressEnter=true to submit.

Press a key or key combination: Enter, Tab, Escape, Backspace, ArrowDown, Control+A, Shift+Enter, F5, etc. Optionally focus an element first.

Choose one or more options in a native or a visible custom combobox by value, label or index. Custom controls are opened and their visible option is clicked automatically; use chrome_cursor when the widget has no semantic combobox/listbox role.

Scroll the page or a container, jump to top/bottom, or bring an element into view. Uses trusted mouse wheel events when CDP is available.

Wait for a selector/text state, a URL, a JavaScript predicate, or a page load state. Use after actions that trigger async updates or navigation.

Execute JavaScript in the page main world and return a JSON value. Use an IIFE for multiple statements, e.g. "(() => ({title: document.title}))()". Powerful: prefer snapshot/DOM when possible.

Query DOM elements and return metadata, HTML, text, attributes, computed styles, rects or counts. Supports open shadow DOM with the ">>>" separator, e.g. "my-app >>> button".

Fill multiple fields in one call and optionally submit. Each field may use ref or selector; type can be "text", "select", "checkbox" or "radio".

Set one or more files on an using absolute paths on the machine running Chrome. Requires CDP and the top frame.

Read console logs, warnings, errors and unhandled rejections captured since page load. Use action="clear" to reset the buffer.

Read fetch/XHR requests captured since page load, including status, duration and errors. Set includeStatic=true to include scripts/images/cached resource timings.

Read, set, delete or clear cookies for the current site using Chrome cookies API. Values are returned to the model: handle secrets carefully.

Inspect recent alert/confirm/prompt dialogs, clear the log, or change how future dialogs are answered. Interception avoids blocking automation.

Drag from one element/point to another using real mouse events when CDP is available, with synthetic pointer/drag events as fallback. Works for pointer-based UIs and many HTML5 drop zones.

Emulate viewport size, device scale, mobile, user agent, geolocation, color scheme, timezone or offline/network conditions. Requires CDP. Use action="reset" to clear overrides.

Show, attach or detach the Chrome debugger for a tab. CDP enables trusted input, full-page screenshots, evaluation, uploads and emulation. Attaching shows a Chrome debug banner.