eyalzh/browser-control-mcp

What it solves

This project provides a secure way for AI assistants to interact with a user's web browser. It allows AI agents to manage tabs, search through browsing history, and read webpage content without granting the AI full, unrestricted access to the browser's internal scripting or page modification capabilities.

How it works

It consists of two components: an MCP (Model Context Protocol) server and a Firefox browser extension. The server acts as the bridge between the AI assistant (like Claude Desktop) and the browser. The extension runs inside Firefox, providing the necessary permissions to access browser data. Communication between the server and the extension is secured via a local-only connection using a shared secret.

Who it’s for

Users who want their AI assistants to perform research, organize their browser tabs, or retrieve information from their browsing history while maintaining strict security and user consent for accessing page content and screenshots.

Highlights

• Secure Access: Unlike typical web automation tools, it does not support arbitrary scripting or page modification.
• Consent-Based Reading: Reading webpage content requires explicit user consent for each domain.
• Controlled Screenshots: Screenshots require individual tab authorization via the extension's toolbar button.
• Tab Management: Tools to open, close, reorder, and group tabs based on AI-driven logic.
• History Search: Ability to search and retrieve specific articles or pages from the browser's history.
• Audit Log: Includes an extension-side audit log for monitoring tool calls.