Agent Browser - Setup Guide¶
Source: vercel-labs/agent-browser (38,590⭐) Category: Browser Automation Languages: Rust + Node.js
Fast native Rust CLI for browser automation, purpose-built for AI agents. Uses Chrome for Testing with accessibility-tree-based interaction (snap + refs) plus traditional CSS selectors. No Playwright or Node.js runtime dependency for the daemon.
Installation¶
# Global install (recommended)
npm install -g agent-browser
agent-browser install # Downloads Chrome for Testing (first time only)
# Or via Homebrew (macOS)
brew install agent-browser
agent-browser install
# Or via Cargo (Rust)
cargo install agent-browser
agent-browser install
# Linux: install system dependencies automatically
agent-browser install --with-deps
# Update
agent-browser upgrade
Prerequisites¶
| Requirement | Details |
|---|---|
| Chrome | agent-browser install downloads from Chrome for Testing. Existing Chrome/Brave/Playwright/Puppeteer installations detected automatically. |
| Node.js 24+ | Only needed when building from source |
| Rust | Only needed when building from source |
| Hermes Agent | Any version - CLI tool, no integration required |
Key Capabilities¶
Core Features¶
| Capability | How to Trigger | Notes |
|---|---|---|
| Open page | agent-browser open <url> |
Launches headless Chrome |
| Accessibility snapshot | agent-browser snapshot |
Returns accessibility tree with refs (@e1, @e2...) |
| Click by ref | agent-browser click @e2 |
Ref-based clicking from snapshot |
| Fill by ref | agent-browser fill @e3 "text" |
Fill input fields by ref |
| Get text | agent-browser get text @e1 |
Extract text content by ref |
| Screenshot | agent-browser screenshot page.png |
Full-page or viewport screenshots |
| CSS selectors | agent-browser click "#submit" |
Traditional selector support |
| JavaScript eval | agent-browser eval "document.title" |
Execute JS in page context |
| Daemon mode | agent-browser daemon |
Persistent browser session |
CLI Command Reference¶
# Quick start flow
agent-browser open example.com
agent-browser snapshot
agent-browser click @e2
agent-browser fill @e3 "test@example.com"
agent-browser get text @e1
agent-browser screenshot page.png
agent-browser close
# Traditional selectors
agent-browser click "#submit"
agent-browser fill "input[name='email']" "user@test.com"
# JavaScript evaluation
agent-browser eval "document.title"
agent-browser eval "JSON.stringify(window.__NEXT_DATA__)"
# Daemon mode for persistent sessions
agent-browser daemon --port 3000
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Web scraping | Use snapshot + get text to extract structured data from any site without building selectors |
| Competitive research | Open competitor sites, screenshot features, extract pricing data |
| Social media monitoring | Automate profile checks, extract follower counts, monitor mentions |
| Form automation | Fill and submit lead gen forms, waitlist signups, directory submissions |
| SEO auditing | Extract meta tags, check page structure, audit accessibility trees |
Troubleshooting¶
| Issue | Fix |
|---|---|
| Chrome not found | Run agent-browser install to download Chrome for Testing |
| Covering element blocks click | Dismiss the covering element (consent banner, modal), take fresh snapshot, retry |
| Linux dependencies missing | Run agent-browser install --with-deps |
| Snapshot too large | Use CSS selectors for targeted interaction instead of full accessibility tree |
| Daemon port in use | Specify alternate port: agent-browser daemon --port 3001 |
Verification¶
# Verify installation
agent-browser --version
# Quick functional test
agent-browser open example.com && agent-browser get text @e1 && agent-browser close
← Skills Catalog | Discovery Page → Powered by CorpusIQ
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.