mcp-browser-kit is an MCP Server whose goal is to enable AI assistants to interact with the user's local browser, and to serve as an alternative to "Claude in Chrome", compatible with any AI assistant or coding agent.
Purpose and positioning: This project treats the local browser as an object that AI assistants can operate, connecting browser capabilities into various MCP clients through the MCP protocol, thereby allowing AI assistants or coding agents to directly perform interactive operations on the browser the user is currently using.
Main capabilities: According to official materials, the server uses stdio transport by default, and also provides an http transport mode. Extensions are divided into two builds, M2 and M3. The M2 build supports more features than the M3 build, so if the browser supports both, prefer M2. The official example usage instruction is: send "Use browser-kit, star the last open GitHub repo on my browser" in the MCP client.
Usage:
- Add MCP Browser Kit to the MCP client, with the stdio configuration: {"mcpServers":{"browser-kit":{"command":"npx","args":["@mcp-browser-kit/server@latest"]}}}.
- To use http transport, run: npx @mcp-browser-kit/server@latest --transport http, with optional parameter defaults of --http-host 127.0.0.1 and --http-port 21082; then switch to using the URL in the MCP client: {"mcpServers":{"browser-kit":{"url":"http://127.0.0.1:21082/mcp"}}}。
- Choose the extension build that matches the browser: the M2 build file names are mcp_browser_kit_m2-<version>.zip for Chromium (for example mcp_browser_kit_m2-5.0.0.zip), and mcp_browser_kit_m2-<version>.xpi for Firefox (for example mcp_browser_kit_m2-5.0.0.xpi); the M3 build file name is mcp_browser_kit_m3-<version>.zip (for example mcp_browser_kit_m3-5.0.0.zip).
- Download the latest compatible build from The Latest Release, load it into a compatible browser and enable it, then refresh already-open tabs to ensure the extension scripts are loaded.
- Send a message in the MCP client to start using it.
Installation methods for each browser extension:
- Chromium-based browsers (Chrome, Brave, Edge): extract the downloaded .zip, open chrome://extensions/, enable Developer mode, click "Load unpacked", and select the extracted extension folder.
- Firefox: open about:addons, click the gear icon and select "Install Add-on From File", and select the downloaded .xpi file.
- Safari: extract the .zip, open Develop > Show Extension Builder, click + and select Add Extension, select the extracted extension folder, and click Install to install.
Applicable scenarios: Suitable for users who want to let AI assistants or coding agents directly operate the local browser in any MCP client, and for users who need an alternative to "Claude in Chrome".
Limitations and security considerations: During the Alpha stage, the official recommendation is to consider using a separate browser profile or a dedicated browser instance to prevent sensitive data from being unintentionally exposed to AI model providers. Currently, incoming requests are not authenticated, so --http-host defaults only to 127.0.0.1 (loopback only); only when you understand the risk that "anyone who can access that host/port can control your browser" should you pass --http-host 0.0.0.0 (or a LAN address). BrowserKit currently lacks authentication for incoming requests, and port 21082 as well as ports 2769-2799 should only be kept within the device's local network. The official service also provides a hosted deployment at Fronteir AI.
