Home

Unbrowse

Open-source API reverse engineering for AI agents.

Unbrowse captures real browser network traffic and turns it into reusable agent skills. Instead of automating browsers with slow, brittle DOM manipulation, agents call the underlying APIs directly — the same endpoints the website's own frontend uses.

Browse once. Teach forever. Every website is an API.

How It Works

1. Capture — Record network traffic from a real browser session
2. Learn — Infer endpoints, auth patterns, and request shapes
3. Generate — Output portable skill artifacts (SKILL.md, scripts, auth config)
4. Replay — Execute captured APIs directly, no browser needed
5. Share — Publish skills so other agents skip the browser entirely

Why?

Browser automation (Playwright, Puppeteer, etc.) forces machines to pretend to be humans — rendering pixels, parsing DOM, clicking buttons — to trigger API calls that could be made directly. Unbrowse short-circuits that path.

• Speed: Direct API calls run in ~200ms vs 10-30s for browser automation
• Reliability: Stable endpoints instead of fragile CSS selectors
• Security: No browser context to exploit (research shows browsing agents have significant vulnerabilities)

Installation

# OpenClaw skill (recommended)
npx skills add unbrowse-ai/unbrowse
openclaw gateway restart

# Standalone (any coding agent)
npx skills add lekt9/unbrowse-openclaw

See Getting Started for detailed setup.

Learn More

• Getting Started — Install, first capture, first replay
• How It Works — Core thesis and capture→learn→replay pipeline
• Skill Format — Anatomy of a generated skill
• Authentication — How auth capture and refresh works
• Marketplace — Publishing and discovering skills
• Research — Academic validation

Links

• GitHub Repository
• unbrowse.ai