Browser Harness MCP Server: Setup for Claude, Cursor and Other Apps

• Opens a new browser tab

  Opens a new tab in your Chrome, optionally at a URL, and returns its tab ID.

  Try: “Open a new tab with example.com.”

  Runs code · browser_new_tab

• Goes to a web address

  Sends the current tab to a URL.

  Try: “Go to the pricing page of this site.”

  Runs code · browser_goto

• Reads the page title and size

  Returns the current tab's URL, title, and viewport and scroll sizes.

  Try: “What page am I on, and how long is it?”

  Read · browser_page_info

• Takes a screenshot of the tab

  Captures a PNG of the current tab and saves it to a file, a temporary one if you give no path. It can shrink large images.

  Try: “Take a screenshot of this page.”

  Files · browser_screenshot

• Lists open tabs

  Lists the page tabs open in your browser.

  Try: “What tabs do I have open?”

  Read · browser_list_tabs

• Shows the active tab

  Returns the active tab's ID, URL and title.

  Try: “Which tab are you looking at?”

  Read · browser_current_tab

• Switches to another tab

  Switches to a tab by its ID or part of its URL.

  Try: “Switch to my Gmail tab.”

  Runs code · browser_switch_tab

• Closes a tab

  Closes a tab by ID, or the active tab if you give none.

  Try: “Close the tab you just opened.”

  Runs code · browser_close_tab

• Moves off internal browser pages

  Switches to a normal web tab if the current one is a chrome:// page or has gone stale.

  Try: “Make sure you are on a real web page.”

  Runs code · browser_ensure_real_tab

• Clicks at a spot on the page

  Clicks at screen coordinates with the left, right or middle button, once or more.

  Try: “Click the Sign in button.”

  Runs code · browser_click

• Types text into the focused field

  Inserts text into whichever element has focus.

  Try: “Type my search phrase into the box.”

  Runs code · browser_type

• Fills in a form field

  Fills an input found by a CSS selector with text, clearing it first by default.

  Try: “Fill the email field with test@example.com.”

  Runs code · browser_fill

• Presses a key

  Presses a key, with optional Alt, Ctrl, Meta or Shift held down.

  Try: “Press Enter.”

  Runs code · browser_press

• Scrolls the page

  Scrolls the mouse wheel at a point by a set number of pixels.

  Try: “Scroll down to the comments.”

  Runs code · browser_scroll

• Waits a few seconds

  Pauses for a set number of seconds.

  Try: “Wait two seconds, then continue.”

  browser_wait

• Waits for the page to finish loading

  Waits until the current tab reports it has finished loading, up to a timeout.

  Try: “Wait for the page to load.”

  Read · browser_wait_for_load

• Waits for an element to appear

  Waits for an element matching a selector to show up, and can require it to be visible.

  Try: “Wait until the results table appears.”

  Read · browser_wait_for_element

• Runs JavaScript in the page

  Evaluates a JavaScript expression in the current tab or in an iframe and returns the result. It runs with your logged-in session.

  Try: “List all link URLs on this page using JavaScript.”

  Runs code · browser_js

• Calls any Chrome DevTools command

  Sends a raw Chrome DevTools Protocol command with your parameters. It can reach nearly anything the browser can do.

  Try: “Use the DevTools protocol to read this tab's cookies.”

  Runs code · browser_cdp

• Attaches a local file to an upload field

  Sets a file from your computer on a file input found by selector.

  Try: “Upload report.pdf to this form.”

  Write · browser_upload_file

• Fetches a URL without the browser

  Makes an HTTP GET request outside the browser and returns the body. You can add headers such as an auth token.

  Try: “Fetch this JSON endpoint and show me the result.”

  Network · browser_http_get

• Starts recording actions

  Starts recording the assistant's browser actions to a local folder.

  Try: “Record what you do from now on.”

  Write · browser_start_recording

• Stops recording

  Stops the active recording and returns the folder it was saved in.

  Try: “Stop recording.”

  Files · browser_stop_recording

Watch out for

• It acts as you on every site you are signed in to

  Pages, email, banking and work tools open in your Chrome are all reachable. Use a separate Chrome profile for assistant work if you can. github.com

• It can run any JavaScript or browser command

  Two tools run page scripts and raw DevTools commands, which can read cookies and change pages. Watch what the assistant runs and do not use it on sensitive tabs. github.com

• It can upload your local files

  One tool attaches a file from your computer to a web form. Approve each upload. github.com

• Recordings and screenshots are saved to disk

  Screenshots and optional recordings are written to local folders. Delete them if they show private pages. github.com

Security advisories