Selenium MCP: Documentation

Connection

npx -y @angiejones/mcp-selenium@latest (stdio). No authorization needed — the server spins up a local browser via Selenium WebDriver. Supported browsers: Chrome, Firefox, Edge and Safari (Safari on macOS only; run sudo safaridriver --enable once before first use). The WebDriver BiDi protocol auto-enables on compatible browsers for richer diagnostics.

Tools (16)

• start_browser, close_session — start and end a browser session.
• navigate — go to a URL.
• interact — click, double-click, right-click, hover over an element.
• send_keys, press_key — type text and press individual keys.
• get_element_text, get_element_attribute — read an element's text and attributes.
• upload_file — upload a file through an input element.
• take_screenshot — screenshot the page.
• execute_script — run arbitrary JavaScript on the page.
• window — list, switch and close windows/tabs.
• frame — switch context between frames.
• alert — handle browser dialogs (accept, dismiss, type text).
• add_cookie, get_cookies/delete_cookie — manage cookies.
• Diagnostics — console, errors and network requests (via WebDriver BiDi).

Resources are also available: browser status and an accessibility-tree snapshot of the page.

Agent flow

"Open the site, log in with a test account and screenshot the dashboard", "fill out the signup form and check the error message text", "run a JS script to scroll to the bottom and collect the data" — classic browser automation with no cloud service.

Note

A community project (MIT, 422★), not officially affiliated with the Selenium project — an independent MCP wrapper around Selenium WebDriver by developer angiejones.

Pricing

Free and open source (MIT); requires locally installed browsers and drivers — no paid dependencies.