BrowserMCP — AI Browser Control via MCP

About

BrowserMCP — AI Browser Control via MCP

Give AI agents control over your browser.

BrowserMCP turns Chrome into an MCP server (Model Context Protocol). Connect AI agents like Claude, Cursor, or any MCP client directly to your browser — no npm, no bridge process, no dependencies.

How it works:
1. Install extension — Load unpacked in chrome://extensions/
2. Start relay — python3 relay.py (Python 3 is pre-installed on macOS/Linux)
3. Connect AI agent — Add MCP config, done
4. Click Start — BrowserMCP icon → Start

35 MCP Tools:

Core: navigate, screenshot, get_dom, click, type_text, extract_text, scroll, get_tabs, switch_tab, close_tab, evaluate, get_page_info, fill_form, wait, press_key, get_links

Advanced: start_recording, stop_recording, playback, detect_forms, auto_fill_form, create_tab, batch_execute, get_console_logs, get_network_requests

Pro (requires license): highlight, wait_for_element, get_interactive_elements, click_by_id, type_by_id, click_text, hover, drag_and_drop, handle_dialog, get_markdown

Why BrowserMCP?

• Zero-Setup — No npm, no Node.js, no pip. Just python3 relay.py
• Your real session — Uses your logged-in tabs, not a clean browser
• 100% local — All processing on your machine. No telemetry, no analytics, no cloud
• Minimal & fast — ~26KB extension + 13KB relay. No 300MB+ browser binaries
• Open standard — Built on the Model Context Protocol (MCP)
• Recording/Playback — Record browser actions and replay them
• Batch Execute — Run multiple tool calls in a single request
• Network Capture — Monitor all network requests in real-time
• Markdown Export — Convert any page to clean structured Markdown

Privacy:

BrowserMCP transfers no data to external servers. The relay listens exclusively on localhost. All browser interaction happens locally via WebSocket on 127.0.0.1. No telemetry, no tracking, no analytics.

Requirements:

• Google Chrome (or Chromium-based browser with Manifest V3 support)
• Python 3.x (pre-installed on macOS/Linux; Windows: small download)
• An MCP-capable AI client (Cursor, Claude Desktop, Hermes, etc.)

Configuration:

Add this to your MCP client config:
{
  "mcpServers": {
    "browser-mcp": {
      "command": "python3",
      "args": ["path/to/relay.py"]
    }
  }
}

License: MIT — Open Source

GitHub: https://github.com/Tabtii/browser-mcp
MCP Protocol: https://modelcontextprotocol.io