WeaveTab V2

The Zero-Setup Local Browser MCP for AI Agents

WeaveTab is a lightweight, local Model Context Protocol (MCP) server that empowers AI agents to seamlessly interact with web browsers using Chrome DevTools Protocol (CDP).

Unlike other solutions, WeaveTab is completely autonomous and zero-setup. It automatically launches a sandboxed browser profile parallel to your daily browsing—meaning you never have to close your browser or mess with debugging ports again.

🌟 Key Features

• Zero-Setup Autonomous Browser: Runs completely parallel to your existing browser via an isolated sandbox. No manual port configuration, no closing your personal browser.
• No Extensions Required: Connects directly via native CDP.
• Semantic Action Map: weave_read returns a parsed, highly compressed DOM semantic map, significantly reducing token consumption.
• Sniper Mode Actions: weave_click and weave_type automatically locate and interact with elements in one step based on their intent or label.
• Local & Secure: No cloud dependencies. Built-in rate limiting and optional Safe Mode.
• Screenshot Support: Optionally use weave_screenshot to capture current tabs, or weave_peek for targeted vision on specific elements/canvas.

🚀 Quick Start (MCP Integration)

The easiest way to use WeaveTab is via npx. Add the following to your MCP client configuration (e.g., Cursor, OpenCode, Claude Desktop).

Cursor / OpenCode

Add this to your IDE's MCP configuration settings:

{
  "mcpServers": {
    "weavetab": {
      "command": "npx",
      "args": ["-y", "@fy2ne/weavetab@latest"]
    }
  }
}

Claude Desktop

Edit your claude_desktop_config.json:

{
  "mcpServers": {
    "weavetab": {
      "command": "npx",
      "args": ["-y", "@fy2ne/weavetab@latest"]
    }
  }
}

🛠️ Tool Arsenal

WeaveTab exposes a rich set of specialized tools for your AI:

• weave_read: Reads the page and returns a semantic action map.
• weave_click: Clicks elements by label, intent, or ID.
• weave_type: Types text into forms securely.
• weave_navigate: Navigates to a target URL.
• weave_scroll: Scrolls up, down, top, or bottom.
• weave_wait: Wait for specific states (navigation, element, DOM stability).
• weave_tabs: List open tabs or switch between them.
• weave_extract: Extract visible content to text or JSON.
• weave_screenshot: Capture a PNG of the current view.
• weave_peek: Targeted vision over specific coordinates.
• weave_key: Press physical keys and modifiers.
• weave_find: Find elements semantically without extracting the full page.

🛡️ Security & Configuration

WeaveTab respects your system. It actively refuses to run as root/sudo.

On first run, WeaveTab generates a configuration file at ~/.weavetab/config.json:

{
  "safeMode": false,
  "allow": [],
  "rateLimitMs": 1000,
  "preferredBrowser": "google-chrome",
  "persistentProfile": false,
  "screenshot": true,
  "peek": true
}

Safety Features

• Storage hard-block: Disables scripts accessing cookies/localStorage in specific environments.
• Root Detection: Exits if run with root permissions.
• Domain Allowlist: Prevents your agent from going to untrusted sites when safeMode is active.
• Audit Logging: Every action taken by the AI is cleanly logged locally at ~/.weavetab/audit.log for your review.

🤝 Contributing

Contributions are welcome!

git clone https://github.com/fy2ne/WeaveTab.git
cd WeaveTab
npm install
npm run build

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.