Agent Browser CLI Control

agent-browser-cli Control

Skill by ara.so — Devtools Skills collection.

agent-browser-cli is a browser control CLI built in Rust that connects to a real Chrome session via a Chrome extension. It enables AI agents to scan tabs, execute JavaScript, manage cookies, capture screenshots, and perform CDP operations while preserving the user's login state and session.

Key difference from Selenium/Playwright: This tool works with an existing Chrome session (not headless), preserving all user login states and cookies. Perfect for AI agents that need to work with authenticated web sessions.

Installation

Via npm (recommended)

npm install -g @sleepinsummer/agent-browser-cli

Via source

git clone https://github.com/sleepinginsummer/agent-browser-cli.git
cd agent-browser-cli
cargo build --release
# Binary will be at ./target/release/agent-browser-cli

Chrome Extension Setup

Critical: The Chrome extension must be loaded for the CLI to work.

1. Download chrome-extensions.zip from the latest release
2. Extract the zip file
3. Open Chrome and navigate to chrome://extensions
4. Enable "Developer mode"
5. Click "Load unpacked extension"
6. Select the extracted tmwd_cdp_bridge directory
7. Ensure at least one normal web page tab is open (not about:blank or chrome://)

The extension will show a small indicator on the right side of the page when connected.

Architecture

agent-browser-cli runs as a daemon service that:

• Listens on port 18765 for Chrome extension WebSocket connections
• Listens on port 18767 for CLI HTTP API calls
• Maintains persistent browser connection to avoid re-initialization overhead

Performance reference (with daemon running):

• Simple page read / JS execution: 40-120ms
• DOM query operations: 270-360ms
• Page monitoring with change summary: 720-880ms

Core Commands

Daemon Management

# Start daemon (auto-starts if not running)
agent-browser-cli start

# Stop daemon
agent-browser-cli stop

# Restart daemon
agent-browser-cli restart

# Check daemon status
agent-browser-cli status

Tab Management

# List all tabs
agent-browser-cli tabs

# Switch to tab by index
agent-browser-cli switch 0

# Open new URL
agent-browser-cli open https://example.com

# Close current tab
agent-browser-cli close

# Get active tab info
agent-browser-cli active

Page Content Extraction

# Scan current page (simplified HTML)
agent-browser-cli scan

# Get text-only content (faster)
agent-browser-cli scan --text-only

# Get full page DOM
agent-browser-cli scan --full

# Scan specific tab
agent-browser-cli scan --tab 0

# Get page with screenshots of visible elements
agent-browser-cli scan --screenshot

JavaScript Execution

# Execute JavaScript and return result
agent-browser-cli exec 'return document.title'

# Execute with page monitoring (detects changes)
agent-browser-cli exec --monitor 'document.querySelector("#search").value = "test"'

# Multi-line JS execution
agent-browser-cli exec 'const btn = document.querySelector("button");
btn.click();
return "clicked";'

# Execute in specific tab
agent-browser-cli exec --tab 0 'return window.location.href'

Screenshots

# Capture full page screenshot
agent-browser-cli screenshot

# Screenshot specific tab
agent-browser-cli screenshot --tab 0

# Output to specific file
agent-browser-cli screenshot --output /path/to/screenshot.png

Cookie Management

# Get all cookies for current page
agent-browser-cli cookies

# Get cookies for specific domain
agent-browser-cli cookies --domain example.com

# Set cookie
agent-browser-cli cookies --set 'name=value; domain=.example.com; path=/'

Configuration

# Change extension WebSocket port (default: 18765)
agent-browser-cli set-extension-port 18766

# View current config
cat ~/.agent-browser-cli/config.json

Real-World Examples

Example 1: Search Automation

# Open search engine
agent-browser-cli open "https://www.google.com"

# Wait a moment for page load, then input search query
sleep 1
agent-browser-cli exec 'document.querySelector("textarea[name=q]").value = "rust programming"'

# Submit search
agent-browser-cli exec --monitor 'document.querySelector("textarea[name=q]").form.submit()'

# Extract search results
agent-browser-cli scan --text-only

Example 2: Form Filling

# Navigate to form page
agent-browser-cli open "https://example.com/contact"

# Fill form fields
agent-browser-cli exec '
const form = {
  name: document.querySelector("#name"),
  email: document.querySelector("#email"),
  message: document.querySelector("#message")
};
form.name.value = "AI Agent";
form.email.value = "agent@example.com";
form.message.value = "Hello from agent-browser-cli";
return "Form filled";
'

# Submit and monitor changes
agent-browser-cli exec --monitor 'document.querySelector("form").submit()'

Example 3: Multi-Tab Workflow

# Get list of all tabs
TABS=$(agent-browser-cli tabs)
echo "$TABS"

# Switch to first tab
agent-browser-cli switch 0

# Get current page title
agent-browser-cli exec 'return document.title'

# Open new tab
agent-browser-cli open "https://example.com"

# Work with new tab
agent-browser-cli scan --text-only

Example 4: Data Extraction with Cookies

# Navigate to authenticated page
agent-browser-cli open "https://example.com/dashboard"

# Extract cookies (user is already logged in)
COOKIES=$(agent-browser-cli cookies --domain example.com)
echo "$COOKIES"

# Extract protected content
agent-browser-cli scan --text-only

# Take screenshot of dashboard
agent-browser-cli screenshot --output dashboard.png

Example 5: Page Monitoring

# Execute action and monitor DOM changes
RESULT=$(agent-browser-cli exec --monitor '
document.querySelector("#load-more").click();
return "Clicked load more";
')

# The result includes a change summary
echo "$RESULT"

# Re-scan to get updated content
agent-browser-cli scan --text-only

Response Format

All commands return JSON:

{
  "ok": true,
  "result": {
    "status": "success",
    "data": "...",
    "metadata": {}
  }
}

Error format:

{
  "ok": false,
  "error": "Error message"
}

Common Patterns

Wait for Element

agent-browser-cli exec '
return new Promise((resolve) => {
  const check = () => {
    const el = document.querySelector("#target");
    if (el) resolve("Found");
    else setTimeout(check, 100);
  };
  check();
});
'

Extract Structured Data

agent-browser-cli exec '
return Array.from(document.querySelectorAll(".item")).map(item => ({
  title: item.querySelector(".title")?.textContent,
  link: item.querySelector("a")?.href,
  price: item.querySelector(".price")?.textContent
}));
'

File Upload

agent-browser-cli exec '
const input = document.querySelector("input[type=file]");
const dt = new DataTransfer();
dt.items.add(new File(["content"], "test.txt"));
input.files = dt.files;
input.dispatchEvent(new Event("change", { bubbles: true }));
return "File set";
'

Dropdown Selection

agent-browser-cli exec '
const select = document.querySelector("select#country");
select.value = "US";
select.dispatchEvent(new Event("change", { bubbles: true }));
return select.value;
'

Troubleshooting

Extension not connecting

1. Verify extension is loaded in chrome://extensions
2. Check extension port matches config: cat ~/.agent-browser-cli/config.json
3. Ensure at least one normal web page is open (not chrome:// or about:)
4. Restart daemon: agent-browser-cli restart

Command timeout

1. Check daemon status: agent-browser-cli status
2. Verify Chrome is running and extension is active
3. Try simpler command first: agent-browser-cli tabs
4. Check logs: tail -f ~/.agent-browser-cli.log

Port conflicts

# Change extension port
agent-browser-cli set-extension-port 18766

# Update extension popup to match new port
# Restart daemon
agent-browser-cli restart

WSL 2 connectivity issues

For WSL 2 users on Windows 11 22H2+:

1. Enable networkingMode=mirrored in .wslconfig
2. Restart WSL
3. Ensure Chrome extension can connect to localhost:18765

Slow performance

• Use --text-only flag for faster content extraction
• Avoid --screenshot unless needed
• Use --monitor only when detecting page changes is necessary
• Consider increasing system resources if Chrome is sluggish

Advanced: CDP Operations

The underlying Chrome extension uses Chrome DevTools Protocol (CDP). While most operations are abstracted by the CLI, you can execute raw CDP commands via JavaScript:

agent-browser-cli exec '
return await chrome.debugger.sendCommand({tabId: chrome.tabs.getCurrent().id}, "Network.getCookies", {});
'

Environment Variables

The CLI uses a config file instead of environment variables:

Config location: ~/.agent-browser-cli/config.json

{
  "extension_port": 18765
}

Logs

Daemon logs are written to:

~/.agent-browser-cli.log

Monitor in real-time:

tail -f ~/.agent-browser-cli.log

Integration with AI Agents

When called by AI coding agents, typical workflows:

1. Check tab state: agent-browser-cli tabs
2. Navigate or switch: agent-browser-cli open <url> or agent-browser-cli switch <index>
3. Extract content: agent-browser-cli scan --text-only
4. Perform action: agent-browser-cli exec --monitor '<js code>'
5. Verify result: agent-browser-cli scan or agent-browser-cli screenshot

The tool is designed for high-frequency calls with minimal overhead (~40-120ms for simple operations).