Browser MCP
Author: BrowserMCP
Description: Browser MCP is an MCP server + Chrome extension that lets AI clients (VS Code, Claude, Cursor, Windsurf) automate your existing Chrome profile locally (fast, private, logged-in, stealthy). This package publishes an MCP server CLI (`mcp-server-browsermcp`) that runs from `dist/index.js` and communicates over WebSocket. Note: the repo README states it currently cannot be built standalone due to missing monorepo `utils`/`types`/shared workspace packages; for usage and setup see https://docs.browsermcp.io and the website https://browsermcp.io. License: Apache-2.0.
Stars: 6.0k
Forks: 464
License: Apache License 2.0
Category: Specialized
Overview
Installation
FAQs
How does Browser MCP avoid bot detection compared to traditional headless browser automation tools like Playwright?
Browser MCP avoids bot detection by attaching to your existing Chrome profile with established browsing history, cached data, and trusted device fingerprints that sites have whitelisted. Traditional headless tools spawn fresh instances with no history, default settings, and WebDriver flags that anti-bot systems scan for, appearing as automated sessions rather than normal human activity.
What are the steps to troubleshoot Browser MCP when npx fails to start the server on macOS with nvm?
Verify npx resolves correctly by running `which npx` in the terminal your MCP client uses at launch. If the path differs from your interactive shell due to nvm's PATH initialization timing, either configure your client to source nvm before starting or hard-code the absolute npx path in your MCP config's command field to bypass shell environment mismatches.
Can Browser MCP and Playwright MCP be used together in the same workflow for both authenticated and CI/CD automation?
Yes. They are separate MCP servers configured independently in your client's MCP config file. You would add both server entries, then prompt your AI assistant to use Browser MCP for local authenticated tasks and Playwright MCP for headless or CI-ready tasks. Each server runs its own process and maintains its own browser context without interfering with the other.
What specific browser automation tools and actions does Browser MCP expose to MCP clients through its API?
Browser MCP exposes navigation, clicking, keypresses, waits, and accessibility snapshots through its tool API. These commands operate through the Chrome extension's WebSocket connection to drive the attached browser profile. The tool surface mirrors common browser automation primitives but executes them in your active Chrome session rather than spawning isolated instances, allowing assistants to interact with already-authenticated pages programmatically.
Why can't the Browser MCP public GitHub repository be built standalone, and how does this affect contributing to the project?
The public repository is a read-only mirror of a private monorepo, meaning it lacks the build infrastructure, shared dependencies, or workspace configuration present in the source repository. This prevents external contributors from compiling or testing changes locally, effectively limiting community contributions to issue reports or documentation suggestions rather than code patches.
How do I handle WebSocket connection drops during long Browser MCP sessions to prevent workflow interruptions?
Click the Connect button in the extension side panel to re-establish the WebSocket connection. Design workflows with logical checkpoints where reconnection won't lose critical state. Structure automation into smaller, resumable segments rather than single long-running tasks, and verify connection health before each major step. Some clients expose connection status that can be monitored programmatically to trigger automatic reconnection attempts.