Parallel Browser Automation: True Per-Tab CDP Concurrency Benchmark

How WebBridge breaks the single-active-tab bottleneck, enabling AI agents to drive multiple Google Chrome tabs concurrently with reproducible benchmarks.

1. The Single-Slot Architecture Bottleneck

In traditional extensions and bridges, CDP commands are sent without a specific tab handle:

• Every command operates on whatever tab happens to be focused in the foreground.
• If an agent attempts to scrape Tab A and Tab B at the same time, the commands interleave, corrupting DOM state and stealing focus.
• Batch scraping 20 pages requires waiting 20 × $T$ seconds sequentially.

2. The WebBridge Per-Tab Map Architecture

In agent-webbridge-extension/src/dbg.mjs, debugger attachments are maintained in a concurrent map:

// Per-tab debugger attachment map in WebBridge MV3 service worker:
const activeSessions = new Map(); // tabId -> { version, attachedAt }

export async function attachTab(tabId) {
  if (activeSessions.has(tabId)) return;
  await chrome.debugger.attach({ tabId }, "1.3");
  activeSessions.set(tabId, { attachedAt: Date.now() });
}

export async function sendCommand(tabId, method, params) {
  await attachTab(tabId);
  return await chrome.debugger.sendCommand({ tabId }, method, params);
}

Because each tab receives independent CDP event streams, operations in Tab 1 never block or interfere with operations in Tab 2.

3. Reproducible Live Benchmark Results

In our automated test suite (scratch/live_m1.mjs), throwaway tabs are spawned and instructed to evaluate an async sleep interval concurrently:

4. Parallel Dispatch Code Example

Here is how to dispatch parallel evaluations across multiple tabs using standard Node.js Promise.all:

import { WebBridgeClient } from './awb-client.mjs';

const client = new WebBridgeClient({ profile: 'Work', session: 'batch-audit' });

// 1. Open 3 parallel tabs
const urls = [
  'https://github.com/trending',
  'https://news.ycombinator.com',
  'https://producthunt.com'
];

const tabIds = await Promise.all(
  urls.map(url => client.navigate(url, { newTab: true }).then(res => res.tabId))
);

// 2. Concurrently extract titles across all tabs in parallel
const titles = await Promise.all(
  tabIds.map(tabId => 
    client.command('evaluate', { 
      _tabId: tabId, 
      code: 'document.title' 
    }).then(r => ({ tabId, title: r.value }))
  )
);

console.log('Extracted in parallel:', titles);

// 3. Clean up the entire session
await client.closeSession();

Frequently Asked Questions