AI AGENT BROWSER

A real browser for agents that need the live web.

Remote Browser gives coding agents and browser-use style workflows a cloud Chromium they can connect to over CDP, watch live, and reuse across authenticated runs.

Create an agent browser Read the agent skill

agent-playwright.tsremote

import { chromium } from "playwright";

const browser = await chromium.connectOverCDP(
  process.env.REMOTE_BROWSER_CDP_URL!
);
const context = browser.contexts()[0] ?? await browser.newContext();
const page = context.pages()[0] ?? await context.newPage();
await page.goto("https://example.com");

CDPnative connection

livehuman viewer

profilesreused login state

Agent-ready CDP

Pass a single WebSocket URL to Playwright, Puppeteer, browser-use, Codex, Claude, Hermes, or your own browser controller.

Visible sessions

Watch what the agent sees, take over when a workflow needs human judgment, and debug failures with screenshots or recordings.

Persistent profiles

Use named browser profiles so agents can continue from a prepared account state instead of repeating login on every run.

Workflow

From browser creation to evidence.

Create the browser

Choose region, timeout, profile, recording, proxy, and launch arguments through the dashboard API or integration plugin.

Connect the agent

The agent receives a CDP URL and drives the session with the same browser actions it already knows.

Observe and close

Use live viewer and recordings for evidence, then terminate the session or let idle timeout protect runtime spend.

FAQ

Short answers for searchers and AI crawlers.

Can a coding agent use Remote Browser directly?

Yes. The Remote Browser skill reads REMOTE_BROWSER_CDP_URL, treats it as secret, and tells the agent to connect through a CDP-capable browser tool instead of launching local Chrome.

Is this only for browser-use?

No. browser-use is one supported workflow. Any tool that can connect over CDP can use the same remote session.

When should I use a remote browser instead of a fetch API?

Use a browser when the task depends on JavaScript, login state, layout, clicks, screenshots, downloads, or human-visible evidence.

Keep moving through the browser automation map.