Agent Browser: A Complete Setup Guide

What You Get After Installing#

Once Agent Browser is set up, your AI assistant can open browsers, click buttons, fill forms, take screenshots, and extract page content through terminal commands. It’s not a tool for writing automation scripts manually — it’s built for AI agents. The AI sees a page snapshot and decides what to click next.

Unlike Browser Use (Python ecosystem), Agent Browser is implemented in Rust and communicates via the Chrome DevTools Protocol. No Node.js runtime needed for browser control. Fast startup, fits naturally into CLI workflows.

Before You Start#

Make sure you have:

• Node.js 18+ (for npm-installing the CLI itself; browser control doesn’t need Node)
• Or Homebrew on macOS, or a Rust environment to install via Cargo
• Roughly 500MB disk space (first install downloads Chrome for Testing)

node -v   # confirm 18+

Installation#

The fastest way is npm global install:

npm install -g agent-browser

After installing, download the browser on first use:

agent-browser install

This automatically downloads Chrome for Testing. If you already have Chrome or Brave installed locally, Agent Browser can detect and reuse them.

On Linux, add a flag to install system dependencies:

agent-browser install --with-deps

macOS users can also use Homebrew:

brew install agent-browser
agent-browser install

Verify the Installation#

Try opening a page:

agent-browser open https://example.com

If the browser launched (or headless mode ran without errors), you’re good. Next, grab a page snapshot:

agent-browser snapshot -i --json

It returns the page’s accessibility tree, with each interactive element tagged with a reference marker (@e1, @e2, etc.). AI agents use these markers to locate elements.

Interact with elements using the markers:

agent-browser click @e2        # click the button tagged @e2
agent-browser fill @e3 "hello"  # type "hello" into the input tagged @e3

Configuration#

Create an agent-browser.json in your project root to customize behavior:

{
  "$schema": "https://agent-browser.dev/schema.json",
  "headed": true,
  "proxy": "http://localhost:8080",
  "profile": "./browser-data"
}

Priority from lowest to highest: user config (~/.agent-browser/config.json) → project config (agent-browser.json) → env vars (AGENT_BROWSER_ prefix) → CLI args.

One Thing to Watch Out For#

The default operation timeout is 25 seconds. If you set it above 30 seconds, the CLI may return EAGAIN errors. If an operation hangs, check your timeout setting before reinstalling the browser.

Windows users also need to close all Chrome windows before using the --profile flag to reuse Chrome login state — otherwise the profile gets locked.

Agent Browser vs Browser Use#

Both work in Claude Code or Cursor. Browser Use has a more mature ecosystem (96k Stars) and larger community. Agent Browser performs better (Rust-native), supports multiple browser engines (Chrome, Safari, iOS simulator, cloud browsers), and includes security features like domain allowlists and operation policies.

Choose Agent Browser if you need iOS Safari control or cloud browser integration. Choose Browser Use if you just need basic web automation with a lower setup barrier.

What’s Next#

Try having your AI agent run a complete task flow — “open this website, log in, extract that data.” Agent Browser supports batch commands and session persistence for chaining complex workflows. Pair it with find-skills to discover more available skills — browser automation is just the start.