browser-element-inspection

Use browser_element with operation="attribute" and the attribute name to read one attribute of an indexed element, or browser_find_elements with a CSS selector and an attributes list to read attributes from multiple elements at once.

Call browser_element with operation="evaluate" and an arrow function like () => this.checked or () => this.disabled. This reads the live DOM state directly instead of inferring it from rendered text, which can be misleading.

browser_evaluate runs a single arrow function with optional args and returns its value, suiting one-liner computations. browser_evaluate_js runs arbitrary browser JavaScript or an IIFE for richer extraction, with max_chars to cap large outputs.

No. Both evaluate tools are bound by the same-origin policy, so they cannot read cross-origin iframes or closed shadow roots. Those reads return nothing or throw; use iframe and shadow DOM specific handling instead.

Run browser_observe with includeAccessibility=true to add the CDP Accessibility tree with roles and accessible names. For a single element, use browser_element evaluate with () => this.getAttribute('role') to read its explicit role attribute.

browser_element uses the index from the latest observation, so a stale index points to the wrong element or nothing after page updates. Re-observe the page to get fresh indices before inspecting elements again.