Report browser running state and list open pages without starting the browser.

Create and select a new page, optionally opening an HTTP/HTTPS URL.

Select the current page by a pageId returned from list_pages.

Close the specified page, or the current page when pageId is omitted.

Open an HTTP/HTTPS URL in the current page and wait for DOMContentLoaded.

Navigate the current page to its previous history entry.

Navigate the current page to its next history entry.

Reload the current page and wait for DOMContentLoaded.

List visible interactive elements in the main document or a frame, with bounded output and temporary references.

Read visible text from the first matching region in the main document or a frame, bounded by maxChars and --max-text-chars.

Capture the viewport, the full page, or a single element as a PNG image. Pass target to crop to one element, which costs far fewer tokens than a full viewport image.

Click an element reference returned by snapshot or a CSS selector. Use frameId with selectors inside a frame. Returns the resulting url, title, navigated, and any newPages.

Move the mouse over an element reference or CSS selector without clicking.

Click page coordinates in CSS pixels from the viewport's top-left corner. Useful for controls inside cross-origin iframes that snapshot cannot enumerate.

Press on an element, move with a human-like curved path, and release. Give exactly one destination: toTarget to drop onto another element, or dx/dy to drag by a pixel offset (what slider controls need). Both ends must be on screen at once. Returns the from and to points actually used.

List frames and iframes with stable frameId values, parent relationships, URLs, and names. Request bounding boxes only when coordinate interaction needs them.

Detect common browser interstitials and embedded cross-origin challenge controls. Returns present, kind, widgetState, tokenPresent, the widget box, and related frames.

Use human-like coordinate clicks on a standard checkbox inside a cross-origin challenge frame, then poll until clearance is confirmed by a response token, widget success state, or interstitial exit. Does not require a vision model. ok means the page is not blocked and you may continue - it does not mean a challenge was defeated. Report what happened from challengeObserved, interactionPerformed, and clearanceVerified: on a page that was never challenged all three are false while ok is true.

Focus an element and enter text, optionally clearing its current value and pressing Enter afterward. Use frameId with selectors inside a frame.

Select one or more native dropdown options by value or label.

Set a checkbox state, or select a radio element with checked=true. Radios cannot be unchecked directly.

Press a key in the current page, such as Enter, Escape, ArrowDown, or Control+A.

Scroll the current page; positive deltaY scrolls down and negative deltaY scrolls up.

Wait for one typed condition: element, text, URL glob, page load state, or duration.