Puppeteer vs Playwright, decided by the job

Same lineage, different scope

Both are JavaScript-first browser automation libraries with very similar shapes: launch a browser, open contexts and pages, find elements, evaluate script in the page. Playwright’s own migration guide maps most Puppeteer calls one to one, for example puppeteer.launch to chromium.launch and createIncognitoBrowserContext to newContext. Moving a script between them is mostly renaming rather than rewriting.

The scope is what differs. Puppeteer describes itself as a reference implementation of browser automation over the DevTools Protocol and WebDriver BiDi, and stays a library. Playwright grew into a cross-engine framework: patched Firefox and WebKit builds, official packages in four more languages, and Playwright Test with fixtures, retries, parallel workers, sharding, tracing and reporters.

That makes the decision mostly about what you are building. A test suite needs everything Playwright adds. A rendering service, a crawler or a one-off script may need none of it, and then a smaller library tied to Chrome’s own protocol is a reasonable preference rather than a compromise.

Scraping with either tool

For scraping, the useful differences are practical. Playwright’s request context sends and stores the browser context’s cookies, so a scraper can sign in through the page and then call the JSON endpoint behind it directly, which is usually faster and more stable than parsing the DOM. Proxies can be set per context in both tools, with credentials in Playwright’s options or through page.authenticate in Puppeteer, so rotating identities within one browser is possible either way.

Rendering fidelity is where Puppeteer has an edge by default. It runs Chrome’s new headless mode, which Chrome’s documentation describes as the real browser without a window. Playwright runs the separate chrome-headless-shell unless you choose the chromium channel, and its own documentation warns that the two can behave differently. If a site renders differently headless, check which headless build you are running before blaming the tool.

Neither project offers anti-bot evasion. The plugin wrappers most often cited for it, puppeteer-extra and playwright-extra, both last published to npm in March 2023, so treat guides built on them with caution. Respect robots.txt, the site’s terms and rate limits; a scraper that behaves politely is cheaper to keep running than one that fights the site it depends on.

Building a test suite on Puppeteer

It can be done. Puppeteer’s FAQ points to community projects such as jest-puppeteer, and Node’s built-in test runner works with no extra dependency. You then own browser lifecycle, a fresh browser context per test, default timeouts, screenshots on failure, retries, parallelism across machines and reporting. Each piece is small; together they are the framework Playwright Test already is.

Synchronisation needs care too. Puppeteer’s locators wait for elements to be visible, enabled and stable and retry the action, but page.$, element handles and page.click with a selector each behave differently, so a suite that mixes styles gets inconsistent waiting. Playwright applies the same actionability checks to every action and retries assertions, which removes that class of bug by default.

The usual path is that a team starts with Puppeteer scripts, adds tests around them, and eventually wants WebKit, traces or sharding. If that is the direction, start in Playwright and avoid the second migration.

Running either in CI and containers

Both download pinned browsers at install time, and both have the same failure mode in modern package managers that block install scripts: the browser never arrives. For Puppeteer, run npx puppeteer browsers install explicitly; for Playwright, npx playwright install with its system dependencies, or start from the official Docker image. Puppeteer 25 also needs Node.js 22.12 or later and an ES module project, which catches older build tooling.

Image size matters for scraping fleets and serverless functions. puppeteer-core downloads nothing, so it suits images that bring their own Chrome. Playwright’s --only-shell option installs just the headless shell when no headed or branded browser is needed. On Ubuntu 23.10 and later, Chrome’s sandbox can be blocked by an AppArmor policy; apply the documented workaround rather than disabling the sandbox, which is the protection against hostile pages that scrapers in particular rely on.