How does ref-based selection work?

It generates an accessibility snapshot where interactive elements are assigned simple identifiers like @e1 or @e2. This allows Claude to interact with elements accurately without needing complex CSS or XPath selectors.

Does it support authenticated sessions?

Yes, you can manage independent browser sessions using the --session flag, allowing you to maintain different login states or perform parallel tasks.

Can I see the browser while it's running?

Yes, by adding the --headed flag to your commands, you can launch a visible Chromium window to watch the automation in real-time and debug interactions.

Is this an alternative to the Playwright MCP?

Yes. While Playwright MCP uses tool-based responses, agent-browser is a CLI-first tool that allows Claude to perform automation directly through Bash commands, which is often faster for one-off tasks.

Agent Browser Automation

Name: Agent Browser Automation
Author: jjeremycai

byjjeremycai

0•

브라우저 자동화

Automates headless web interactions and data extraction using an LLM-friendly CLI with ref-based element selection.

Agent Browser is a specialized CLI tool designed to give AI agents a reliable way to interact with the web. By utilizing accessibility snapshots and a unique reference-based selection system (@e1, @e2), it simplifies complex DOM interactions into stable Bash commands that Claude can execute with high precision. Whether you need to navigate multi-step login flows, fill out intricate forms, or capture high-quality screenshots, this skill provides a robust alternative to traditional scripting by optimizing the interface specifically for the Claude Code workflow.

주요 기능

01Ref-based element selection using accessibility snapshots

020 GitHub stars

03Interactive element filtering to optimize AI context windows

04Semantic locators for finding elements by role, text, or label

05Support for parallel browser sessions with independent states

06Full-page and viewport screenshot and PDF generation

사용 사례

01Automated web scraping and data collection from dynamic, JavaScript-heavy sites

02Capturing visual snapshots of web pages for UI audits or documentation

03End-to-end testing of web applications and automated form submission

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jjeremycai/claudekit agent-browser

For use in Claude.ai and ChatGPT

Download Skill