AgentGo Stealth Browser: Cloud Playwright Automation
AgentGo provides a distributed cloud browser cluster. You connect via WebSocket using chromium.connect() from playwright@1.51.0.
“Stealth” refers to configuration intended to reduce automation detection. Results depend on the target site and cannot be guaranteed.
Important: Pin Playwright to 1.51.0. Newer versions may be protocol-incompatible with AgentGo’s server.
When to use / When not to use
Use this skill when you want:
- A remote browser (cloud) instead of running Chromium locally
- Reliable automation for navigation, form filling, screenshots, testing, scraping/extraction
- Parallel multi-page workflows
Don’t use this skill when:
- A simple HTTP fetch is enough (no JS / no interactions)
- You can’t pin Playwright to
1.51.0
Prerequisites
- Node.js environment to run Playwright code
- An AgentGo account (free credits available)
- Environment variable:
AGENTGO_API_KEY
Get an API key
Register at https://app.agentgo.live/
export AGENTGO_API_KEY="your_api_key_here"
Install (must pin Playwright)
npm install playwright@1.51.0
# or
pnpm add playwright@1.51.0
# Optional: for session management
npm install @agentgo-dev/sdk
Minimal example (open → screenshot → save)
This saves a full-page screenshot to example.png in your current working directory.
import { chromium } from "playwright"; // must be playwright@1.51.0
const options = { _apikey: process.env.AGENTGO_API_KEY };
const serverUrl = `wss://app.browsers.live?launch-options=${encodeURIComponent(
JSON.stringify(options)
)}`;
const browser = await chromium.connect(serverUrl);
try {
const page = await browser.newPage();
await page.goto("https://example.com", { waitUntil: "domcontentloaded" });
await page.screenshot({ path: "example.png", fullPage: true });
} finally {
await browser.close();
}
Quick start
import { chromium } from "playwright"; // must be playwright@1.51.0
const options = { _apikey: process.env.AGENTGO_API_KEY };
const serverUrl = `wss://app.browsers.live?launch-options=${encodeURIComponent(
JSON.stringify(options)
)}`;
const browser = await chromium.connect(serverUrl);
try {
const page = await browser.newPage();
await page.goto("https://example.com");
console.log(await page.title());
} finally {
await browser.close();
}
Browser sessions (recommended)
If you need better control over lifecycle, concurrency, or reuse, use AgentGo browser sessions.
Docs:
Typical flow:
- Create a browser session using the official SDK (
@agentgo-dev/sdk)
- The SDK returns a
connectionUrl (WebSocket)
- Connect with Playwright using that URL
- Run automation, then close the browser to end the session
Session connect example using the SDK:
import { AgentGo } from "@agentgo-dev/sdk";
import { chromium } from "playwright"; // must be playwright@1.51.0
const client = new AgentGo({ apiKey: process.env.AGENTGO_API_KEY! });
const session = await client.sessions.create({ region: "us" });
const browser = await chromium.connect(session.connectionUrl);
try {
const page = await browser.newPage();
await page.goto("https://example.com", { waitUntil: "domcontentloaded" });
await page.screenshot({ path: "session-example.png", fullPage: true });
} finally {
await browser.close();
}
Note: Session creation/usage details can change on the AgentGo side. Always follow the two official docs above for the latest parameters and lifecycle steps.
Connection helper (recommended)
The examples below use this helper. Copy it into your project, or inline the connection logic.
import { chromium } from "playwright";
export async function connectAgentGo() {
if (!process.env.AGENTGO_API_KEY) {
throw new Error("AGENTGO_API_KEY is not set");
}
const opts = encodeURIComponent(
JSON.stringify({ _apikey: process.env.AGENTGO_API_KEY })
);
return chromium.connect(`wss://app.browsers.live?launch-options=${opts}`);
}
Pass _disable_proxy: true in launch options to bypass the default proxy. See session-management.md for all connection options.
Basic interactions
const browser = await connectAgentGo(); // see "Connection helper" section above
const page = await browser.newPage();
await page.goto("https://example.com");
await page.click("button#submit");
await page.fill("input[name=email]", "user@example.com"); // for anti-detection, use keyboard.type() instead — see tips-general.md
await page.press("input[name=email]", "Enter");
await page.screenshot({ path: "screenshot.png" });
await browser.close();
Extract data
const browser = await connectAgentGo(); // see "Connection helper" section above
const page = await browser.newPage();
await page.goto("https://news.ycombinator.com");
const items = await page.$$eval(".titleline a", els =>
els.map(a => ({
title: a.textContent,
href: a.href,
}))
);
await browser.close();
console.log(items);
Multiple pages (parallel)
const browser = await connectAgentGo(); // see "Connection helper" section above
const [page1, page2] = await Promise.all([browser.newPage(), browser.newPage()]);
await Promise.all([
page1.goto("https://site-a.com"),
page2.goto("https://site-b.com"),
]);
await browser.close();
Always close in finally
const browser = await connectAgentGo(); // see "Connection helper" section above
try {
const page = await browser.newPage();
await doWork(page);
} finally {
await browser.close();
}
References (deep dive)
