Tabrix

Releases · npm package · MIT License · TypeScript 5.8+

Turn your real Chrome session into an MCP browser runtime for AI assistants.

Tabrix is a Chrome extension plus a local native server. It lets MCP clients operate the browser profile you already use, including your open tabs, cookies, extensions, and logged-in pages.

Why Tabrix

Most browser automation starts from a fresh browser. Tabrix starts from the browser where your work is already happening.

• Real session reuse: keep the login state, cookies, extensions, and tabs you already rely on.

• MCP-native control: connect through Streamable HTTP or stdio from Codex, Claude Desktop, Cursor, Windsurf, Dify, and similar MCP clients.

• Local-first by default: browser state stays on your machine unless you explicitly enable remote access.

• Operational checks included: tabrix status, doctor, smoke, and report help diagnose the real browser path.

• Object-aware direction: Tabrix is evolving toward page-object awareness, operation memory, reusable knowledge, and AI-reviewed experience records.

Related MCP server: monkeysee

Quick Start

Install the CLI:

npm install -g @tabrix/tabrix@latest
# or
pnpm install -g @tabrix/tabrix@latest

Register the native host if needed:

tabrix register

Load the extension:

1. Download the latest extension zip from GitHub Releases.

2. Unzip it into a stable local directory.

3. Open chrome://extensions, enable Developer mode, and load the unpacked extension folder.

4. Open the Tabrix popup and click Connect.

Verify the runtime:

tabrix status
tabrix doctor --fix
tabrix smoke

Connect an MCP client:

{
  "mcpServers": {
    "tabrix": {
      "type": "streamableHttp",
      "url": "http://127.0.0.1:12306/mcp"
    }
  }
}

For detailed setup, use Quickstart and CLI and MCP Configuration.

What You Can Do

• list and switch tabs in your current Chrome session

• read the current page and return structured browser context

• click, fill, type, upload files, and take screenshots

• inspect console, network, history, bookmarks, and browser status

• share the same browser capability with token-protected LAN access

• build AI workflows that reuse real browser context instead of rebuilding it

The complete tool catalog is in API Reference.

Product Direction

Tabrix is not trying to become a browser IDE, workflow SaaS, or embedded AI chat product. The public product center is:

real Chrome + MCP control + browser object awareness + memory and reuse

The core long-term loop is:

• Object: what the current browser page contains and what looks actionable

• Memory: what the AI asked Tabrix to do, what Tabrix returned, and how the AI evaluated the result

• Knowledge: reusable page/source capability summaries distilled from Object records

• Experience: reusable paths distilled from AI-reviewed successful Memory records

Chrome APIs are the default browser execution path. CDP is treated as a bounded sensing and diagnostics enhancer, not as the default public control surface.

Read the public direction in Roadmap and capability status in Product Surface Matrix.

Documentation

Users

• Quickstart

• CLI and MCP Configuration

• Troubleshooting

• Use Cases

Developers

• Contributing

• Architecture

• Project Structure

• Testing

Reference

• Docs Index

• API Reference

• Compatibility Matrix

• Platform Support

• Security

• Changelog

Contributing

Start with Contributing. High-impact areas include cross-platform install quality, browser-runtime reliability, MCP compatibility, tool contract consistency, and regression coverage.

Origin

Tabrix is a community-driven continuation of hangwin/mcp-chrome.

License

MIT. See LICENSE.