Screenshots

Take viewport and full-page screenshots from a cloud browser, and control where they’re saved.

A cloud browser is a normal CDP endpoint, so screenshots work the way your framework takes them, and they save wherever your code runs.

Where screenshots are saved

The most-asked question first: screenshots taken through Playwright or Puppeteer are written by your code, to a path you choose. Nothing is stored on the browser — the browser’s automatic recording is separate.

Python

from playwright.async_api import async_playwright
from browser_use import BrowserUse

client = BrowserUse()
browser = client.browsers.create()

async with async_playwright() as p:
    pw = await p.chromium.connect_over_cdp(browser.cdp_url)
    page = pw.contexts[0].pages[0]
    await page.goto("https://example.com")
    await page.screenshot(path="shots/example.png")  # your machine, your path

client.browsers.stop(browser.id)

import { chromium } from "playwright";
import { BrowserUse } from "browser-use";

const client = new BrowserUse();
const browser = await client.browsers.create();

const pw = await chromium.connectOverCDP(browser.cdp_url);
const page = pw.contexts()[0].pages()[0];
await page.goto("https://example.com");
await page.screenshot({ path: "shots/example.png" });

await client.browsers.stop(browser.id);

Full page, not just the viewport

By default a screenshot captures the visible viewport. For the whole page, top to bottom:

await page.screenshot(path="full.png", full_page=True)

Playwright stitches the scroll automatically. The result contains page content only, no URL bar or browser chrome, because CDP screenshots capture the rendered page, not the window.

Resolution

Screenshot dimensions follow the browser’s screen size, set at browser creation with screen_width and screen_height. Set them explicitly if screenshots must match a target resolution:

browser = client.browsers.create(screen_width=1920, screen_height=1080)

Screenshots vs recording

Screenshots are moments; recording is the whole browser session as video — every browser records automatically, and you fetch a fresh MP4 URL on demand. For debugging agent behavior, recording is usually what you want; for artifacts and QA evidence, screenshots.

From agent tasks

Ask the agent to take screenshots as part of a task and collect them from the run’s output files (GET /runs/{id}/files?kind=output).