What problem does it solve?
AI agents often struggle to interact with live websites using standard automation tools that are detectable or require heavy dependencies. agent-browser lets AI agents drive your real Chrome via CDP, preserving your cookies, running headed, and avoiding redundant relays.
Core Features & Use Cases
- Direct CDP-based control of your actual Chrome browser (headed or headless) with a single WebSocket.
- Real-session isolation: separate cookies, localStorage, and tabs per agent/session; supports parallel tasks.
- AI-native element references (@eN) and compact snapshots to dramatically reduce prompt size and improve reliability.
- Open-world browsing tasks: login, form fill, navigation, data extraction, screenshot capture, network interception, and JavaScript execution on live sites.
- Use cases include verifying deployments, testing dashboards (Google Ads, Stripe, Cloudflare), and automating repetitive browser tasks.
Quick Start
Install agent-browser, run start-chrome-debug to launch a persistent Chrome, then use agent-browser to open a site, snapshot the UI to obtain @e refs, and interact using those refs.