Give your IDE a browser driver’s license

What it does

browser-use-mcp-server is a protocol adapter that wraps the browser-use automation library in an MCP layer so AI agents inside editors like Cursor, Windsurf, or Claude Desktop can drive a live Chromium session through Playwright. You type a request—“open Hacker News and return the top article”—and the agent translates it into actual clicks, scrolls, and page reads.

The interesting bit

Most MCP tools return text; this one returns a live browser session. The built-in VNC streaming means you can watch the AI navigate pages or fumble through forms in real time, which is either reassuring or deeply unnerving. It also speaks both SSE and stdio, so it works as a local sidecar or a networked service.

Key highlights

• Exposes Playwright-driven browser automation to any MCP-compatible client
• Dual transport support: SSE for remote connections, stdio for local editor integration
• Real-time VNC streaming on port 5900 to observe automation visually
• Async task execution so long-running browser jobs don’t block the client
• Docker image included with a bundled VNC server for isolated deployments

Caveats

• Requires an OpenAI API key; the README doesn’t document alternative LLM providers
• stdio mode depends on an additional mcp-proxy tool rather than speaking stdio natively
• Default VNC password is literally browser-use, so you’ll want to override that in any real environment

Verdict

Worth a look if you want your coding assistant to browse documentation, submit forms, or run web workflows without leaving the editor. Skip it if you just need headless scraping—this is agentic browsing with all the ceremony that implies.