Agent-Browser – Browser Automation for AI
8 hours ago
- agent-browser is a browser automation CLI designed for AI agents, featuring compact text output to minimize token usage.
- It uses a ref-based accessibility tree for deterministic element selection, requiring only 200-400 tokens per snapshot.
- Supports 50+ commands for navigation, forms, screenshots, network, storage, tabs, frames, and debugging.
- Includes built-in observability tools: video recording, streaming, debugging, profiler, and diffing.
- First-class support for modern apps with network control, React, Web Vitals, Next.js, Vercel, and event extension workflows.
- Stateful capabilities: sessions, profiles, auth state, cookies, storage, proxy, and security controls for long-running agents.
- Cross-platform: macOS, Linux, Windows with native binaries; works with AI coding agents like Claude Code, Cursor, GitHub Copilot.
- Client-daemon architecture: Rust CLI parses commands and communicates via Chrome DevTools Protocol; daemon persists and auto-exits after idle timeout.