Agent Browser

You can run a browser automation MCP server in the cloud to perform browser actions from Cursor Agents or a local Cursor. This server exposes endpoints to open pages, interact with elements, take screenshots, and manage sessions, all through MCP-compatible commands.

How to use

Connect your MCP client to the remote agent browser MCP server at the provided endpoint. Use the server's capabilities to perform real browser actions: open a URL and start a session, click elements, fill forms, type text, wait for elements, take screenshots, and close the session when you are finished. Authentication tokens are managed per session and persist for subsequent actions within the same session.

Available tools

browser_open

Opens a URL and starts a new browser session. Supports basic authentication when needed.

browser_snapshot

Obtains an accessibility snapshot of the current page.

browser_click

Clicks a specified element on the page.

browser_fill

Fills a form field with a provided value.

browser_type

Types text into a field, simulating human-like input.

browser_get_text

Retrieves visible text from a page element.

browser_screenshot

Captures a screenshot of the current page.

browser_wait

Waits for a specified element to appear before continuing.

browser_press_key

Sends a keystroke to the page.

browser_close

Closes the current browser session.