Remote Chrome

Control a Chrome browser running on the host via the Chrome DevTools Protocol (CDP).

Prerequisites

1. Host Setup (User Must Do This)

Chrome must be started with remote debugging enabled on the host:

# macOS
/Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222

# Linux
google-chrome --remote-debugging-port=9222

# Windows
chrome.exe --remote-debugging-port=9222

# Headless mode (no visible window)
google-chrome --headless --remote-debugging-port=9222 --no-sandbox --remote-allow-origins=*

2. Port Forwarding (User Must Do This)

Forward port 9222 into the container:

# From host terminal (outside container)
boxctl ports forward 9222

Or add to .boxctl.yml:

ports:
  container:
    - '9222'

Verifying Connection

Always verify the connection before attempting automation:

# Check if CDP is reachable
curl -s http://localhost:9222/json/version

# List available targets (tabs/pages)
curl -s http://localhost:9222/json/list

Expected response from /json/version:

{
  "Browser": "Chrome/XXX.X.X.X",
  "webSocketDebuggerUrl": "ws://localhost:9222/devtools/browser/UUID"
}

Connection Methods

Method 1: Node.js with chrome-remote-interface (Recommended)

# Install the package
npm install chrome-remote-interface

const CDP = require('chrome-remote-interface');

async function main() {
    const client = await CDP({ port: 9222 });
    const { Page, Runtime, DOM } = client;

    // Enable required domains
    await Promise.all([Page.enable(), Runtime.enable(), DOM.enable()]);

    // Navigate to a URL
    await Page.navigate({ url: 'https://example.com' });
    await Page.loadEventFired();

    // Execute JavaScript in the page
    const result = await Runtime.evaluate({
        expression: 'document.title'
    });
    console.log('Page title:', result.result.value);

    // Get page HTML
    const { root } = await DOM.getDocument();
    const { outerHTML } = await DOM.getOuterHTML({ nodeId: root.nodeId });
    console.log('HTML length:', outerHTML.length);

    await client.close();
}

main().catch(console.error);

Method 2: Python with pychrome

pip install pychrome

import pychrome

# Connect to Chrome
browser = pychrome.Browser(url="http://localhost:9222")

# Get first tab or create new one
tabs = browser.list_tab()
if tabs:
    tab = tabs[0]
else:
    tab = browser.new_tab()

tab.start()

# Enable required domains
tab.Page.enable()
tab.Runtime.enable()
tab.DOM.enable()

# Navigate
tab.Page.navigate(url="https://example.com")
tab.wait(5)  # Wait for load

# Execute JavaScript
result = tab.Runtime.evaluate(expression="document.title")
print("Title:", result['result']['value'])

# Get HTML
doc = tab.DOM.getDocument()
html = tab.DOM.getOuterHTML(nodeId=doc['root']['nodeId'])
print("HTML length:", len(html['outerHTML']))

tab.stop()

Method 3: Direct WebSocket (Low-Level)

import json
import websocket

# Get WebSocket URL
import requests
version = requests.get('http://localhost:9222/json/version').json()
ws_url = version['webSocketDebuggerUrl']

# Connect
ws = websocket.create_connection(ws_url)

# Send CDP command
def send_command(method, params=None, cmd_id=1):
    msg = {"id": cmd_id, "method": method}
    if params:
        msg["params"] = params
    ws.send(json.dumps(msg))
    return json.loads(ws.recv())

# Example: Get browser version
result = send_command("Browser.getVersion")
print(result)

ws.close()

Best Practices

Helper Functions

Use these helpers in every CDP script:

const fs = require('fs');

// Screenshot helper
async function screenshot(Page, name) {
    const { data } = await Page.captureScreenshot({ format: 'png' });
    const path = `/tmp/chrome-${name}.png`;
    fs.writeFileSync(path, Buffer.from(data, 'base64'));
    return path;
}

// Sleep helper
const sleep = ms => new Promise(r => setTimeout(r, ms));

// Click at coordinates
async function clickAt(Input, x, y) {
    await Input.dispatchMouseEvent({ type: 'mousePressed', x, y, button: 'left', clickCount: 1 });
    await Input.dispatchMouseEvent({ type: 'mouseReleased', x, y, button: 'left', clickCount: 1 });
    await sleep(500);
}

// Click button by text - ALWAYS query fresh positions, never cache
async function clickButton(Runtime, Input, buttonText) {
    const pos = await Runtime.evaluate({
        expression: `
            const btn = Array.from(document.querySelectorAll('button'))
                .find(b => b.textContent.trim() === '${buttonText}');
            if (btn) {
                const rect = btn.getBoundingClientRect();
                JSON.stringify({ x: rect.x + rect.width/2, y: rect.y + rect.height/2 });
            }
        `,
        returnByValue: true
    });
    if (pos.result.value) {
        const {x, y} = JSON.parse(pos.result.value);
        await clickAt(Input, x, y);
        return true;
    }
    return false;
}

// Check for blocking overlays (modals, dialogs, popups)
async function hasBlockingOverlay(Runtime) {
    const result = await Runtime.evaluate({
        expression: `
            const selectors = '.modal, [role="dialog"], .overlay, .popup, .dialog';
            const overlays = document.querySelectorAll(selectors);
            Array.from(overlays).some(el => {
                const style = window.getComputedStyle(el);
                const rect = el.getBoundingClientRect();
                return style.display !== 'none'
                    && style.visibility !== 'hidden'
                    && rect.width > 0;
            });
        `,
        returnByValue: true
    });
    return result.result.value === true;
}

// Try multiple strategies to close overlays
async function closeAnyOverlay(Runtime, Input) {
    // Strategy 1: Press Escape
    await Input.dispatchKeyEvent({ type: 'keyDown', key: 'Escape', code: 'Escape', windowsVirtualKeyCode: 27 });
    await Input.dispatchKeyEvent({ type: 'keyUp', key: 'Escape', code: 'Escape', windowsVirtualKeyCode: 27 });
    await sleep(300);
    if (!await hasBlockingOverlay(Runtime)) return true;

    // Strategy 2: Click Close/Cancel button
    if (await clickButton(Runtime, Input, 'Close')) {
        await sleep(300);
        if (!await hasBlockingOverlay(Runtime)) return true;
    }
    if (await clickButton(Runtime, Input, 'Cancel')) {
        await sleep(300);
        if (!await hasBlockingOverlay(Runtime)) return true;
    }

    // Strategy 3: Click X button (common patterns)
    const xBtn = await Runtime.evaluate({
        expression: `
            const selectors = '.close, .btn-close, [aria-label="Close"], button:has(svg)';
            const btn = document.querySelector('.modal ' + selectors + ', [role="dialog"] ' + selectors);
            if (btn) {
                const rect = btn.getBoundingClientRect();
                JSON.stringify({ x: rect.x + rect.width/2, y: rect.y + rect.height/2 });
            }
        `,
        returnByValue: true
    });
    if (xBtn.result.value) {
        const {x, y} = JSON.parse(xBtn.result.value);
        await clickAt(Input, x, y);
        await sleep(300);
        if (!await hasBlockingOverlay(Runtime)) return true;
    }

    return false; // Could not close
}

Verify-Act-Verify Pattern

Never assume an action worked. Always:

1. Screenshot BEFORE to see current state
2. Check for blocking overlays and close them
3. Perform the action
4. Screenshot AFTER
5. READ the screenshot to visually confirm success

async function safeClick(Page, Runtime, Input, buttonText, stepName) {
    // 1. Screenshot before
    await screenshot(Page, `${stepName}-1-before`);

    // 2. Clear any blocking overlays
    if (await hasBlockingOverlay(Runtime)) {
        console.log('Overlay detected, closing...');
        await closeAnyOverlay(Runtime, Input);
        await screenshot(Page, `${stepName}-2-overlay-closed`);
    }

    // 3. Perform action
    console.log(`Clicking ${buttonText}...`);
    const clicked = await clickButton(Runtime, Input, buttonText);
    if (!clicked) {
        console.log(`Button "${buttonText}" not found!`);
        return false;
    }

    // 4. Screenshot after - MUST READ THIS to verify
    const path = await screenshot(Page, `${stepName}-3-after`);
    console.log(`Verify result: ${path}`);
    return true;
}

// Usage
await safeClick(Page, Runtime, Input, 'Submit', 'submit-form');
// Then READ /tmp/chrome-submit-form-3-after.png to verify it worked

Why This Matters

| Problem | Solution | |---------|----------| | Modals block clicks | Check hasBlockingOverlay() before every action | | Cached coordinates are stale | Always query DOM fresh with clickButton() | | Can't tell if action worked | Screenshot after and READ it | | Different apps close differently | closeAnyOverlay() tries Escape, Close, Cancel, X |

Common Mistakes

Bad: Click then assume success

await clickAt(Input, 500, 300);  // Hope it works

Good: Verify before and after

await screenshot(Page, 'before');
if (await hasBlockingOverlay(Runtime)) {
    await closeAnyOverlay(Runtime, Input);
}
await clickButton(Runtime, Input, 'Submit');
await screenshot(Page, 'after');  // READ THIS FILE

Common Operations

Navigate and Get Content

// Navigate
await Page.navigate({ url: 'https://example.com' });
await Page.loadEventFired();

// Get page content
const { root } = await DOM.getDocument();
const { outerHTML } = await DOM.getOuterHTML({ nodeId: root.nodeId });

Click Element

// Find element by selector
const { root } = await DOM.getDocument();
const { nodeId } = await DOM.querySelector({
    nodeId: root.nodeId,
    selector: '#my-button'
});

// Get element position
const { model } = await DOM.getBoxModel({ nodeId });
const x = model.content[0] + model.width / 2;
const y = model.content[1] + model.height / 2;

// Click
await Input.dispatchMouseEvent({ type: 'mousePressed', x, y, button: 'left', clickCount: 1 });
await Input.dispatchMouseEvent({ type: 'mouseReleased', x, y, button: 'left', clickCount: 1 });

Type Text

// Focus input first
await Runtime.evaluate({
    expression: `document.querySelector('#my-input').focus()`
});

// Type text
for (const char of 'Hello World') {
    await Input.dispatchKeyEvent({ type: 'keyDown', text: char });
    await Input.dispatchKeyEvent({ type: 'keyUp', text: char });
}

Take Screenshot

const { data } = await Page.captureScreenshot({ format: 'png' });
require('fs').writeFileSync('screenshot.png', Buffer.from(data, 'base64'));

Execute JavaScript and Get Result

const { result } = await Runtime.evaluate({
    expression: `
        // Your JavaScript here
        document.querySelectorAll('a').length
    `,
    returnByValue: true
});
console.log('Link count:', result.value);

Troubleshooting

"Connection refused" on port 9222

1. Verify Chrome is running with --remote-debugging-port=9222 on host
2. Verify port is forwarded: boxctl ports forward 9222
3. Check from container: curl http://localhost:9222/json/version

"No targets available"

Chrome has no open tabs. Open a tab in Chrome or create one via CDP:

curl http://localhost:9222/json/new?about:blank

WebSocket connection fails

Use the exact webSocketDebuggerUrl from /json/version response. Do not construct it manually.

Multiple clients

CDP supports multiple clients since Chrome 63. If another client disconnects you, you'll receive a detached event with reason replaced_with_devtools.

Security Warning

Remote debugging gives full control of the browser. Anyone who can connect to port 9222 can:

• Read all page content including passwords
• Execute JavaScript in any page
• Access cookies and session data
• Navigate to any URL

Only enable on trusted networks and ensure firewall blocks external access to port 9222.

Key CDP Domains

| Domain | Purpose | |--------|---------| | Page | Navigation, screenshots, lifecycle | | Runtime | JavaScript evaluation | | DOM | Document structure, queries | | Input | Mouse and keyboard events | | Network | Request interception, caching | | Emulation | Device simulation, geolocation | | Target | Tab management |

Full protocol documentation: https://chromedevtools.github.io/devtools-protocol/