List all open Chrome page tabs with their targetIds, titles, and URLs. When you are unsure which tab to inspect, call this proactively to discover available tabs, then present the list to the user and ask which one to switch to — do NOT tell the user to switch tabs manually in Chrome.

Switch the MCP connection to a specific Chrome tab. Use list_tabs first to get available targetIds.

Get the title of the currently connected tab (document.title).

Get the URL of the currently connected tab (location.href).

Get the full HTML source of the currently connected tab (document.documentElement.outerHTML). Truncated to 20000 characters for large pages; a truncated result ends with a … marker giving how much was cut and the real length of the document, so a short result is never ambiguous between "small page" and "cut off here".

Evaluate a JavaScript expression in the page, in global scope, with the same semantics as the DevTools console. Returns the real value when it serialises; objects that cannot (DOM nodes, Errors, Maps, class instances) come back as a preview instead: class name plus a first level of properties, marked … where Chrome truncated it — readable, not parseable as the value. Top-level await works, but an expression that merely RETURNS a promise is NOT awaited — it comes back as a pending Promise, exactly as in the console. The call returns as soon as your expression finishes its synchronous work, before queued microtasks run, so the state triggered by a click is not visible in the same call: put await Promise.resolve() between the click and the read, or read in a second call. At a breakpoint this still evaluates globally and cannot see local or closure variables — use evaluate_at_frame for those. For the element selected in the Elements panel ($0), use get_inspected_element.

Get computed CSS values for the given properties on the element matched by selector.

Capture a PNG screenshot of the current viewport (the visible page area only — not the full scrollable page, not the browser chrome, not DevTools), or of one rectangle of it with region. Captured at the tab's native pixel size unless you cap it with maxEdge; a capture that was scaled or cut says so in a note beside the image, which for a region also gives the CSS rect the image covers and how many image pixels a CSS pixel became. A tab that is not painting is raised in its window first, which changes which tab is selected there.

Get the element marked for MCP inspection. To mark an element: select it in the Elements panel, then run window.$0 = $0 in the DevTools console.

Get current debugger state: whether execution is paused, the pause reason, hit breakpoints, and the full call stack with file/line info.

Inspect variable values in a call frame scope. Only works when execution is paused. Use get_debugger_state first to find available frame indices.

Set a breakpoint by URL (exact or regex) + line number.

Remove a breakpoint by ID. The ID is invalidated; use set_breakpoint to restore (returns a new ID).

List breakpoints tracked by this server (set via set_breakpoint). Breakpoints set outside this server (DevTools UI, other CDP clients, prior sessions) are not visible — CDP has no API to enumerate them.

Pause JavaScript execution immediately. After pausing, use get_debugger_state to inspect the call stack.

Resume JavaScript execution after a breakpoint or pause.

Execute the current line and pause at the next line (does not enter function calls). Returns the new call stack position.

Step into the function call on the current line. Returns the new call stack position.

Evaluate a JavaScript expression in the scope of a paused call frame — it reads local variables, closure variables and the current this, which evaluate_js cannot. Only works while execution is paused: when it is not, this returns an error rather than silently falling back to global scope. Results come back as a preview — class name plus a first level of properties, marked … where Chrome truncated it — readable, not parseable as the value. Use get_debugger_state to find frame indices.

Step out of the current function and pause at the caller. Returns the new call stack position.

Return browser console messages and uncaught exceptions. Includes messages already visible in DevTools before this server connected, plus new output produced afterwards. Exceptions are reported with their full stack trace (source-mapped when available).

Return HTTP requests captured from the connected tab — method, URL, resource type, status, transferred size, duration, initiator, and failure reason. Capture starts when this server connects to the tab: requests issued before that are NOT visible (unlike get_console_logs, which replays pre-connect history). Requests belonging to a previous page are pruned on navigation, mirroring the DevTools Network panel default; the new document request itself is kept. A redirect chain appears as one record per hop, sharing a requestId and distinguished by hop. WebSocket frames are not captured.

Fetch the response body for one requestId from get_network_requests. Bodies are never buffered by this server — they are read from Chrome on demand, and Chrome discards them on navigation or when its own buffer limits are exceeded, so fetch promptly and before navigating away. Chrome stores at most one body per requestId, so for a redirect chain only the final hop has a body. Returns a JSON metadata block, then the body as a separate text block — kept separate so a large body is not JSON-escaped. Metadata carries requestId, base64Encoded (boolean; true when the body is base64, as CDP reports it), byteLength and method/url/status/mimeType, the last four replaced by a note when the capture buffer no longer holds the request, plus truncated: true when the body was cut at 50000 characters. A binary body is never returned: metadata carries omitted and there is no second content block.

Discard the buffers this server holds, so a following get_console_logs / get_network_requests shows only what happens next. Affects this server only: nothing is cleared in Chrome or in the DevTools UI, and capture keeps running — no reconnect or reload is needed. Cannot be undone. Console entries are gone for good, because Console.enable() replays history only at attach time. Cleared network requestIds still resolve in get_network_response_body for as long as Chrome itself holds the body; once Chrome drops it the error says the data was discarded rather than that the id is unknown. Works without a connected tab.