obra/superpowers-chrome

What it solves

Superpowers Chrome provides a way for AI agents (like Claude Code) to directly control a web browser via the Chrome DevTools Protocol (CDP). It eliminates the need for complex setup or heavy dependencies, allowing agents to navigate pages, interact with elements, and extract data from the web in a real-time, programmatic way.

How it works

The project implements two distinct modes of operation:

1. Skill Mode: A CLI tool specifically designed for Claude Code agents, providing 17 distinct commands for browser automation (e.g., navigate, click, fill, extract).
2. MCP Mode: A lightweight Model Context Protocol (MCP) server that provides a single use_browser tool for MCP-compatible clients. This mode includes "auto-capture" features that automatically save HTML, Markdown, and screenshots after every DOM-changing action.

It uses a simplified tab index syntax (e.g., 0, 1, 2) to target tabs instead of complex WebSocket URLs. It also handles browser dialogs (alerts, prompts, and permission requests) by surfacing them as synthetic responses, allowing agents to interact with them using a specialized selector grammar (dialog::accept, dialog::dismiss).

Who it’s for

• AI Agent Developers: Those building tools for LLMs to browse the web.
• Claude Code Users: Users who want to give their AI agent browsing capabilities.
• MCP Client Users: Users of Claude Desktop or other MCP-compatible software.

Highlights

• Zero Dependencies: Built-in WebSocket support, requiring no npm install for the core CLI tool.
• ** لقب Idiotproof API**: Simplified tab management using indices.
• Dialog Handling: Specialized system to prevent browser dialogs from blocking agent actions.
• Platform Agnostic: Works across macOS, Linux, and Windows.
• Auto-Capture: In MCP mode, automatically generates DOM summaries, screenshots, and Markdown of the page state after interactions.
• Full CDP Access: Includes a raw command for advanced users to provide full Chrome DevTools Protocol access.