agent-browser

agent-browser is a browser automation CLI built specifically for AI agents, delivering compact accessibility-tree output that uses far fewer tokens than JSON or full DOM representations. It features a native Rust CLI for instant command parsing, a Node.js daemon managing Playwright under the hood, and an experimental pure-Rust native daemon via direct CDP. With 50+ commands covering navigation, forms, screenshots, network, and storage, it integrates seamlessly with any agent that can run shell commands.

• Agent-first output — Snapshot returns a compact accessibility tree with unique refs (e.g., @e1), using ~200–400 tokens vs ~3,000–5,000 for full DOM, making it ideal for LLM context efficiency.
• Ref-based element selection — Each snapshot element gets a deterministic ref, enabling precise, no-re-query interactions like agent-browser click @e2.
• Fast Rust CLI — Native binaries for macOS (ARM64/x64), Linux (ARM64/x64), and Windows (x64) ensure instant command parsing.
• 50+ commands — Full coverage of navigation, forms, screenshots, network inspection, and storage management.
• Sessions — Run multiple isolated browser instances with separate authentication contexts simultaneously.
• CDP Mode — Direct Chrome DevTools Protocol access for advanced use cases.
• Streaming & Diffing — Stream browser output and diff snapshots to track page changes efficiently.
• Profiler — Built-in profiling to measure and optimize agent browser interactions.
• iOS Simulator support — Extend automation to iOS Simulator environments.
• Multiple engine support — Works with Chrome and Lightpanda browser engines.
• Works with leading AI agents — Compatible with Claude Code, Cursor, GitHub Copilot, OpenAI Codex, Google Gemini, opencode, and any shell-capable agent.
• Cross-platform — Install via npm globally, Homebrew on macOS, or try instantly with npx agent-browser.