Automates web browser interactions using a reference-based accessibility tree optimized for AI agents.
Agent Browser is a headless browser automation CLI designed to bridge the gap between AI agents and the web. It leverages a unique accessibility snapshot pattern that allows Claude to interact with web elements via simplified references (like @e1 or @e2) instead of brittle CSS selectors. This skill is ideal for developers needing to scrape data, automate form submissions, or perform end-to-end testing directly from the command line, providing a reliable and bash-friendly workflow for complex web automation tasks.
Key Features
01Multi-session support for managing parallel browser instances
02Structured JSON output for easy data extraction and parsing
03Headless and headed modes for automation and visual debugging
04Reference-based element interaction using accessibility snapshots
05Comprehensive interaction suite including click, fill, scroll, and hover
060 GitHub stars
Use Cases
01Performing web scraping on JavaScript-heavy websites that require interaction
02Automating complex multi-step form submissions and login flows
03Generating screenshots and PDFs of web pages for automated reporting
