Development · GitHub repo
agent-browser
Browser automation CLI and engine purpose-built for AI agents
agent-browser is an open-source browser automation CLI developed by Vercel Labs designed specifically for LLM agents. It provides fast automation with token-efficient accessibility snapshot diffing, visual regression testing, and persistent Chrome DevTools Protocol connections. It integrates with tool ecosystems like Claude Code and remote browser providers.
What it's for
- Verify web application states using token-efficient accessibility snapshot diffs
- Run autonomous dogfooding and QA testing sweeps on deployed websites
- Detect visual UI regressions with pixel-level diffing
- Control headless or headed browser sessions via AI coding assistants like Claude Code
Features
- Accessibility tree snapshot and visual pixel diffing commands
- Persistent CDP session connections and direct remote provider support
- Video recording to WebM and base64 screenshot capture
- Native CLI execution optimized for speed and low token usage
- Claude Code marketplace plugin and WebMCP tool support
- Dogfooding skill for autonomous site audits and issue reporting
- Proxy authentication and custom CA certificate handling
Pricing
Open source. Free (Apache-2.0 license)
Details
- Type
- GitHub repo
- Category
- Development
- Topic
- Browser Automation
- Open source
- Yes
- Platforms
- Command line, macOS, Linux, Windows
- For
- Developers and AI engineers building web-interacting agents and automated QA workflows
- Alternative to
- Playwright, Puppeteer, Browser Use
- Tags
- automation, browser, ai agents, cli, testing, qa, cdp
Why it was saved
- @Shpigford: Recommended by Josh Pigford as substantially faster than Playwright and clearly optimized for AI/agent use.
- @ctatedev: v0.6.0 adds video recording, persistent CDP sessions, proxy auth, computed styles, multi-value select and a Claude Code plugin — a practical tool for building agent/browser workflows.
- @ctatedev: Automated QA/UX auditing of any deployed site using an agent instead of manual testing.
- @ctatedev: Lets you verify page actions with up to 90% fewer tokens, catch unintended CSS/layout regressions before shipping, monitor page changes, compare across environments, and pair with bisect to pinpoint when a regression was introduced.
- @ctatedev: Diffing lets an agent verify page actions with up to 90% fewer tokens, catch visual regressions, monitor page changes, and compare across environments — via snapshot (accessibility-tree) diffs or pixel-level visual diffs.
Compare
Related
Cloud-hosted stealth browser infrastructure and execution platform for AI agents.
Open-source web apps and AI browser tools built with Hyperbrowser
Lightweight Go binary exposing an HTTP API to control Chrome browsers
Run browser automation agents inside on-demand isolated sandbox environments
Self-improving browser automation skill for AI coding agents
AI computer-use agent that learns and automates browser tasks by demonstration
Updated