Browsermcp MCP

What browsermcp/mcp MCP server does

The browsermcp/mcp MCP server connects AI applications to a user's browser through MCP and a Chrome extension. It is intended for browser automation performed in the browser the user already operates, rather than creating a separate browser instance. The repository contains the core MCP code for Browser MCP.

This design is useful when browser state matters. The extension works with the existing browser profile, so services where the user is already signed in remain available to the automation. The browser's existing fingerprint is also retained, which the project says helps avoid basic bot detection and CAPTCHAs.

How it works

An AI client communicates with the MCP server, while the Chrome extension provides access to the user's browser. Automation is performed locally on the user's machine. According to the project, this keeps browser activity on the device instead of sending it to remote servers and avoids network latency during automation.

The browsermcp/mcp MCP server was adapted from Microsoft's Playwright MCP server. The key difference described by the project is that Browser MCP automates the user's active browser context rather than opening new browser instances. This allows the automation to use the user's browser profile and logged-in sessions.

The README identifies VS Code, Claude, Cursor, and Windsurf as AI applications that can use Browser MCP. It does not provide a list of individual MCP tool names in the supplied material.

Tools and capabilities

The browsermcp/mcp MCP server supports these documented capabilities:

• Automate a browser through AI applications.
• Run browser automation locally.
• Use the user's existing browser profile.
• Preserve access to services where the user is already logged in.
• Retain the user's real browser fingerprint.
• Avoid basic bot detection and CAPTCHAs, as described by the project.

These capabilities make the project a fit for agent workflows that must interact with an already configured browser session. They do not establish support for arbitrary browsers, a remote hosted mode, or a particular set of page-interaction tools.

Limitations and notes

The browsermcp/mcp MCP server is not currently buildable from this repository by itself. The README says it depends on utilities and types from the monorepo where it is developed. The supplied material therefore does not provide a standalone build or installation command for this repository.

The project describes a Chrome extension as part of the system, but the excerpt does not specify its installation steps, required Chrome version, MCP client configuration, permissions, or available tool schemas. It also does not state whether browser automation can be run without the extension.

Local execution keeps activity on the user's device, but using an existing browser profile means the automation can operate within that profile's authenticated sessions. Developers should account for the access and security implications of granting an AI application control over that browser context.