Agent Infra Browser

@agent-infra/browser is dedicated to building a comprehensive browser infrastructure SDK specifically designed for AI Agents.

What is this for?

This toolkit is specifically designed for:

• GUI AI Agent that needs to interact with web browsers
• Browser screen casting in non-VNC or headless scenarios
• MCP service for browser automation control

Architecture

Packages Overview

@agent-infra/browser

Core Browser Control Library. Abstracts and encapsulates the fundamental capabilities required to manipulate browsers.

@agent-infra/browser-ui

Browser Screen Casting UI Components. Can connect to remote browsers via CDP and then display their screen casting content.

@agent-infra/browser-finder

Cross-Platform Browser Detection. Automatically locate installed browsers (Chrome, Edge, Firefox) on Windows, macOS, and Linux systems.

@agent-infra/browser-context

Smart Web Content Extraction. Extract clean, readable content from web pages and convert to Markdown format with advanced algorithms and browser automation support.

@agent-infra/media-utils

Media Processing Utilities. Media tools for handling browser-related tasks, such as high-performance base64 image parsing and media resource processing.

Development

This is a monorepo managed with pnpm. To get started:

# Install dependencies
pnpm install

# Build all packages
pnpm run build

# Run tests
pnpm run test

# Lint code
pnpm run format

Requirements

• Node.js >= 20.x
• pnpm for package management
• Chrome/Chromium browser for browser automation features

License

This project is licensed under the Apache License 2.0.

Credits

Special thanks to the open source projects that inspired this toolkit:

• puppeteer - The underlying browser automation library
• Chrome DevTools Protocol - Chrome DevTools Protocol
• ChatWise - Browser detection functionality reference
• readability - A standalone version of the readability lib
• edge-paths - Edge browser path finder