Steel Documentation
Steel gives AI agents real browsers in the cloud. Your agent uses websites
the way a person does, and you can see everything it did.
Start in one step
Choose who drives the browser. The first step is right here.
Read https://setup.steel.dev/install.md and follow it to set up Steel for this coding client. Then continue my requested task using the installed Steel skill.
Approve the sign-in link that the agent gives you, then give it a task, such as "Use Steel to open a browser, send me the live view link, and wait until I say I'm watching. Then open https://steel.dev/practice/news, list the top 3 stories, and stop the session."
Set up your agentclaude mcp add --transport http steel https://mcp.steel.dev/mcp
Run in your terminal, then use /mcp in Claude Code to sign in.
Install the SDK and Playwright, set STEEL_API_KEY, and run:
npm install steel-sdk playwright
import Steel from "steel-sdk";
import { chromium } from "playwright";
const client = new Steel();
const session = await client.sessions.create();
try {
const browser = await chromium.connectOverCDP(session.websocketUrl);
const context = browser.contexts()[0];
const page = context.pages()[0] ?? (await context.newPage());
await page.goto("https://steel.dev/practice/news");
console.log(await page.title());
await browser.close();
} finally {
await client.sessions.release(session.id);
}Already use Browser Use, Playwright or Puppeteer? Each guide runs a first program on Steel that you watch live, then moves the code that you have.
Write Steel codeChoose how to control a browser
Control interfaces compares them in full.
- Web scrapingMarkdown, screenshots and PDFs
- Playwright or Puppeteer codeConnect over CDP
- An agent frameworkBrowser Use, Stagehand, OpenAI Agents SDK
- A model that reads and actsSnapshots with refs, step by step
- A computer-use modelScreenshots, clicks and keys
- An agent with a shell, or a terminalClaude Code, Codex, Cursor