@browser-control/mcp-server

Control Chrome and Edge with AI agents through Model Context Protocol (MCP).

Website: https://browser-control.arjun.tools · GitHub: https://github.com/arjun-g/browser-control

🔐 Flagship Use Case: Automate Inside Your Real, Logged-In Sessions

Browser Control drives the browser you're already signed into — not a headless copy, not a service account. That means it can operate inside admin panels, internal dashboards, and SSO/2FA-protected apps that ordinary scrapers and headless bots cannot reach safely.

• No credentials, API keys, or passwords are ever shared with the AI client or MCP server.
• No re-login, no session replay, no cookie exporting — the agent acts through the tab you already have open.
• Ideal for internal tools, admin consoles, and back-office workflows that require a real authenticated session.

Example prompt:

In our admin dashboard, find all users who signed up this week, export the list, and confirm the export succeeded with a screenshot.

What It Does

This package is the MCP server half of Browser Control Bridge. It connects your MCP client (Claude, GitHub Copilot, Cursor, Continue.dev, and others) to the Browser Control Bridge extension over a local WebSocket bridge, so your agent can:

• Navigate pages and manage tabs
• Click, type, scroll, and interact with UI elements
• Read and snapshot DOM content
• Run JavaScript and inject CSS
• Capture screenshots
• Inspect performance metrics and Core Web Vitals
• Use Chrome DevTools Protocol (CDP) commands

Powerful Features

• Seamless MCP integration with existing browser sessions
• 40+ browser automation tools
• Full DOM interaction and extraction
• Performance and debugging insights
• Tab and window management
• CDP support for advanced workflows

Best Use Cases

• Automate authenticated browser tasks inside admin panels and internal tools (flagship use case — see above).
• Smoke-test critical web flows such as signup, login, checkout, or contact forms.
• Reproduce frontend bugs and attach evidence like screenshots, DOM output, and console logs.
• Extract structured data from dynamic pages that require real rendering and interaction.
• Run lightweight performance and UX checks with Web Vitals, console output, and viewport changes.
• Verify downloads and generated artifacts from browser-based workflows.

Featured Demo Workflow

▶️ Watch the demo: Browser Control — Demo (YouTube)

The demo shows the agent operating a real, already-signed-in Cloudflare Web Analytics dashboard — reading every key metric for the last 24 hours, switching the date range to the last 7 days, and returning a side-by-side comparison with screenshots.

Example prompt:

In the browser tab, read all the key metrics for the last 24 hours — visits, page views, load time, Core Web Vitals, top countries, referrers, browsers, and devices. Then switch the date range to the last 7 days and read the same metrics. Give me a side-by-side comparison highlighting what changed, and screenshot both views.

Get Started

1. Install Extension

Install from the Chrome Web Store — the same listing also works in Microsoft Edge. A native Edge Add-ons listing is still under review.

2. Start MCP Server

Run via npx:

npx -y @browser-control/mcp-server

Or install globally and run the binary:

npm install -g @browser-control/mcp-server
browser-control-mcp

Optional flags/env:

• --agent-name or -a to set agent name
• BROWSER_BRIDGE_PORT to set preferred bridge port
• BROWSER_BRIDGE_TOKEN to require token auth
• BROWSER_AGENT_NAME as fallback agent name

3. Configure Your AI Client

Common MCP config:

{
  "mcpServers": {
    "browser-control": {
      "command": "npx",
      "args": ["-y", "@browser-control/mcp-server"]
    }
  }
}

Claude Code

claude mcp add browser-control -- npx -y @browser-control/mcp-server

GitHub Copilot (VS Code)

Add the MCP server in VS Code MCP settings (or .vscode/settings.json) and enable Agent mode in Copilot Chat.

Cursor

Use Settings -> Features -> MCP -> Add New MCP Server:

• Name: browser-control
• Command: npx
• Args: -y @browser-control/mcp-server

Continue.dev

Add the same MCP config to ~/.continue/config.json.

MCP Tools

Navigation and Tabs

• navigate
• back
• forward
• new_tab
• close_tab
• close_tabs
• reload_tab
• list_tabs
• get_current_tab
• create_tab_group
• delete_tab_group

Interaction

• click
• type
• keypress
• scroll
• mouse_move
• drag_and_drop
• focus_element
• blur_element
• scroll_element

DOM and Content

• read_dom
• dom_snapshot
• minimal_snapshot
• semantic_snapshot
• dom_extract_element
• screenshot
• wait_for_selector
• wait_for_navigation

JavaScript and Styling

• execute_javascript
• add_css

Browser State

• get_cookies
• set_cookie
• delete_cookie
• set_viewport
• resize_window
• emulate_mobile
• toggle_fullscreen

Analytics and Debugging

• get_performance_metrics
• get_web_vitals
• get_console_logs
• list_downloads
• cdp_command

How It Works

• MCP server (this package): Node.js process exposing browser tools over MCP
• Browser extension: Manifest V3 extension running in Chrome/Edge
• WebSocket bridge: local connection between server and extension

Everything runs locally on your machine — no data leaves it.

Support and Links

• Website: https://browser-control.arjun.tools
• GitHub: https://github.com/arjun-g/browser-control
• Issues: https://github.com/arjun-g/browser-control/issues
• Privacy Policy: https://browser-control.arjun.tools/privacy-policy