Can I run multiple browser sessions at once?

Yes, the skill supports a --session flag which allows you to manage and interact with multiple independent, parallel browser instances.

Is it possible to see the browser window while it runs?

Yes, you can use the --headed flag with your commands to launch a visible browser window, which is helpful for debugging visual layouts.

How does ref-based selection work?

It generates unique, short tags (like @e1) for interactive elements within an accessibility snapshot, ensuring the AI targets the correct element without needing to guess complex CSS selectors.

How does Agent Browser differ from Playwright MCP?

Agent Browser uses a CLI-first approach with ref-based selection from accessibility snapshots, making it highly optimized for terminal-based Bash workflows compared to the tool-based Playwright MCP.

Do I need to install a browser separately?

No, you can simply run 'agent-browser install' after installing the npm package to download the necessary Chromium binaries automatically.

Agent Browser CLI Automation

Name: Agent Browser CLI Automation
Author: gvkhosla

bygvkhosla

•

浏览器自动化

Automates headless browser interactions through a streamlined CLI using ref-based element selection optimized for AI agents.

Agent Browser is a specialized CLI tool designed by Vercel specifically for AI-driven web automation within terminal environments like Claude Code. It simplifies complex web interactions by providing accessibility snapshots with reference tags (e.g., @e1, @e2), allowing Claude to accurately click, fill forms, and scrape data without managing complex CSS selectors. This skill is ideal for developers who need a robust alternative to Playwright MCP that integrates directly into Bash workflows for tasks ranging from automated testing to dynamic data extraction.

主要功能

01Semantic locators for finding elements by role, text, or label

02Ref-based element selection for reliable AI interactions

0335 GitHub stars

04Snapshot-driven workflow using accessibility trees

05Full-page and viewport screenshot/PDF generation

06Session management for parallel browser instances

使用场景

01Automating multi-step form submissions and user registration flows

02Scraping dynamic content from SPAs using interactive element snapshots

03Capturing visual documentation and screenshots of web interfaces

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gvkhosla/compound-engineering-pi agent-browser

For use in Claude.ai and ChatGPT

主要功能

01Semantic locators for finding elements by role, text, or label

02Ref-based element selection for reliable AI interactions

0335 GitHub stars

04Snapshot-driven workflow using accessibility trees

05Full-page and viewport screenshot/PDF generation

06Session management for parallel browser instances

使用场景

01Automating multi-step form submissions and user registration flows

02Scraping dynamic content from SPAs using interactive element snapshots

03Capturing visual documentation and screenshots of web interfaces

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gvkhosla/compound-engineering-pi agent-browser

For use in Claude.ai and ChatGPT