Accessibility Scanner MCP Server: Setup for Claude, Cursor and Other Apps

• Checks the open page for WCAG accessibility problems with axe

  Runs the axe-core rule set on the current page and lists each violation with its impact, affected elements and a fix link. It can also save a screenshot with the problems boxed.

  Try: “Scan the page you just opened for WCAG AA problems.”

  Network · scan_page

• Crawls pages on the same site and totals the accessibility problems

  Follows internal links from the current page, scans each page and saves a JSON report that combines the violations found across the site.

  Try: “Audit the whole docs site and give me the worst pages.”

  Write · audit_site

• Compares scan results across screen sizes, color schemes and zoom levels

  Runs the same accessibility scan under several variants, such as mobile and desktop, dark mode, reduced motion and zoom, and compares each to the baseline. Saves a JSON report.

  Try: “Compare accessibility on this page at mobile and desktop sizes.”

  Write · scan_page_matrix

• Tabs through the page to check focus order, visible focus and skip links

  Presses Tab through the page and reports focus jumps, missing focus outlines, skip links and keyboard traps. Saves a JSON report.

  Try: “Check whether I can reach every control on this page with only the keyboard.”

  Write · audit_keyboard

• Checks element names and reading order the way a screen reader would see them

  Reads the browser accessibility tree and element positions to flag poor or missing names and odd reading order. It does not run an actual screen reader. Saves a JSON report.

  Try: “Check whether the buttons on this page have useful names.”

  Write · audit_screen_reader

• Opens a web address in the browser

  Loads the URL you give it in the controlled browser, including local development servers.

  Try: “Open http://localhost:3000 in the browser.”

  Network · browser_navigate

• Goes back to the previous page

  Moves the browser one step back in its history.

  Try: “Go back one page.”

  Read · browser_navigate_back

• Reads the page as a list of controls and text

  Captures the accessibility tree of the page. The assistant uses it to see what is on screen and which element to act on.

  Try: “Show me what is on this page.”

  Read · browser_snapshot

• Searches the page snapshot for words or a pattern

  Looks through the snapshot for text or a regular expression and returns matching elements with a few lines of context.

  Try: “Find the "Sign up" button on the page.”

  Read · browser_find

• Clicks an element on the page

  Clicks the element the assistant picks from the snapshot. Clicks can submit forms or trigger actions on the live site.

  Try: “Click the "Continue" button.”

  Write · browser_click

• Types text into a field

  Enters text into an editable element, with an option to press Enter after.

  Try: “Type my test email into the email field.”

  Write · browser_type

• Fills several form fields in one step

  Sets multiple fields at once, such as text boxes, checkboxes, radios and dropdowns.

  Try: “Fill in the signup form with test data.”

  Write · browser_fill_form

• Presses a keyboard key

  Sends one key press such as Tab, Enter or an arrow to the page.

  Try: “Press Escape to close the menu.”

  Write · browser_press_key

• Picks an option in a dropdown

  Selects one or more options in a select element.

  Try: “Choose "United Kingdom" in the country dropdown.”

  Write · browser_select_option

• Moves the pointer over an element

  Hovers over an element so menus and tooltips appear.

  Try: “Hover over the account menu.”

  Read · browser_hover

• Drags one element onto another

  Performs drag and drop between two elements on the page.

  Try: “Drag the first card onto the done column.”

  Write · browser_drag

• Drops files or data onto an element as if dragged in from outside

  Simulates dragging files or clipboard-like data from outside the browser onto an element. Files are read from your computer unless upload folders are restricted.

  Try: “Drop test.csv onto the upload area.”

  Write · browser_drop

• Uploads files from your computer to the page

  Attaches one or more local files to a file chooser on the page. Any path on your machine works unless you restrict upload folders.

  Try: “Upload ./sample.pdf in the file chooser.”

  Write · browser_file_upload

• Runs JavaScript on the page or on one element

  Executes a JavaScript expression inside the loaded page and returns the result. It can read or change anything the page itself could.

  Try: “Run a script to list all images without alt text.”

  Runs code · browser_evaluate

• Accepts or dismisses a browser pop-up

  Answers alert, confirm and prompt dialogs that block the page.

  Try: “Dismiss the confirmation pop-up.”

  Write · browser_handle_dialog

• Saves a picture of the page or one element

  Takes a PNG, JPEG or WebP screenshot and saves it in the output folder. It is for viewing only, not for choosing elements to click.

  Try: “Take a screenshot of the checkout page.”

  Write · browser_take_screenshot

• Waits for text to appear or disappear, or for some seconds

  Pauses until text shows up or goes away, or a set time passes, so slow pages finish loading before the next step.

  Try: “Wait until the "Loading" text is gone.”

  Read · browser_wait_for

• Returns the page's console messages

  Lists console logs, warnings and errors the page has produced.

  Try: “Show any errors from the browser console.”

  Read · browser_console_messages

• Lists the requests the page made

  Returns network requests since the last navigation, with the option to include up to 200 earlier ones from the same tab.

  Try: “List the failed network requests on this page.”

  Read · browser_network_requests

• Shows details of one network request

  Returns the headers (with credentials redacted) and body details for a single request from the list.

  Try: “Show me the response details for request 12.”

  Read · browser_network_request

• Changes the browser window size

  Resizes the window so you can check how a page looks at mobile or desktop widths.

  Try: “Resize the window to 375 by 812.”

  Write · browser_resize

• Switches dark mode, print view or reduced motion

  Sets CSS media features for the page, such as color scheme or reduced motion, and keeps omitted ones unchanged.

  Try: “Turn on dark mode and reduced motion.”

  Write · browser_emulate_media

• Closes the page

  Closes the current page in the browser.

  Try: “Close the current page.”

  Write · browser_close

• Lists, opens, closes or switches browser tabs

  Manages tabs so a multi-page flow can keep several pages open at once.

  Try: “Open a new tab and switch to it.”

  Write · browser_tabs

• Sets how long page loads may take

  Changes the timeout for navigation and page loads on existing tabs. It does not persist if the browser context is recreated.

  Try: “Allow 60 seconds for page loads.”

  Write · browser_navigation_timeout

• Sets how long clicks and other actions may take

  Changes the default timeout for browser actions on existing tabs. It does not persist if the browser context is recreated.

  Try: “Give actions 10 seconds before timing out.”

  Write · browser_default_timeout

• Opens a second, separate browser with its own cookies

  Creates another browser context with its own tabs, cookies and storage, and returns an id the other browser tools accept.

  Try: “Open a separate browser session for the admin login.”

  Write · browser_session_open

• Closes a browser session opened earlier

  Closes an extra session and frees its browser. The default session cannot be closed this way.

  Try: “Close the admin browser session.”

  Write · browser_session_close

Watch out for

• It drives a real browser that can click and submit

  Click, type and form tools act on live pages, so pointing it at a site you are logged in to can have real effects. The README setup uses an isolated profile, which starts with no logins. github.com

• It can run JavaScript in any page it opens

  The evaluate tool runs scripts on the loaded page. Block it with the blocked-tools option if you do not need it. github.com

• It can upload any file from your computer

  The upload and drop tools read local files and send them to the page. Set the allowed upload folders option to limit this to a safe folder. github.com

• Reports and screenshots are saved to disk

  The audit tools and screenshot tool write files to the output folder you set. Existing files are never overwritten by the screenshot and PDF tools. github.com

• Page-defined tools can appear automatically

  If a page exposes its own tools, they are added to the list and sit outside the allow and block lists. Treat their names and results as untrusted. github.com

• Browser download is opt-in

  Installing a browser is off unless you start the server with the install capability. The tool downloads files without checksum checks, per the README. github.com

Security advisories