AgentDeskAI/browser-tools-mcp

What it solves

BrowserTools MCP allows AI coding agents to interact with and inspect a user's active Chrome session. Unlike traditional automation tools that launch a fresh, unauthenticated browser instance, this project enables agents to see exactly what the user sees, including pages where the user is already logged in, avoiding the need to manually recreate authentication states.

How it works

The system consists of two components: an MCP (Model Context Protocol) server and a Chrome extension. The extension attaches to the active browser session and streams telemetry—such as console logs, network activity, and screenshots—to the the MCP server. The server then exposes these as tools and resources to an MCP-compatible AI client (like Cursor or Claude Code), allowing the agent to query the browser's state in real-time.

Who it’s for

Developers using AI coding agents who need to debug live web applications, perform accessibility or SEO audits, and inspect network traffic without losing their current session state.

Highlights

• Live Session Access: Accesses the real browser profile and active tabs, maintaining user logins and authentication.
• Lighthouse Integration: Provides tools to run performance, accessibility, SEO, and best-practices audits directly via the agent.
• Context Window Management: Large data payloads (like full logs or HAR files) are provided as MCP resources rather than inlined, preventing context window overflow.
• Security-First Design: Features loopback-only binding, per-run authentication tokens, and automatic scrubbing of sensitive credentials (JWTs, cookies, etc.) from captured data.
• Multitab Support: Tracks telemetry separately for every tab with DevTools open, allowing agents to target specific tabs by ID.