Stagehand

Stagehand is an SDK for AI browser agents that blends Playwright-style methods with natural-language instructions so an agent can navigate, click, fill forms and extract data from any website. It is maintained by Browserbase and released on GitHub under the MIT License.

What is Stagehand used for?

Playwright was built for testing; Stagehand is built for agents. It exposes act, observe and extract primitives: you describe the action ("upvote the top story") or the data schema you want and the SDK works it out on the page. It ships for TypeScript, Python and Go and still gives you classic goto, click, locator and screenshot methods.

How do Stagehand's self-healing actions work?

When a site changes and a selector breaks, Stagehand detects it and recomputes how to perform the action, so your automation survives redesigns. It also trims the accessibility tree so the model only sees what it needs, which means fewer tokens per step. In the project's own benchmark, one extraction uses about 7.4k tokens versus 35.7k with Playwright.

Is Stagehand faster than Playwright?

According to the project, yes. The runtime runs as an in-browser extension next to the page, cutting round-trip latency: the site shows 4.7 s versus 9.0 s for the same task. It also adds WebMCP support, clipboard access, batched commands, deep locators for nested iframes and Shadow DOM, domain allowlists and blocklists, and built-in OpenTelemetry tracing.

Stagehand pros and cons

Pros: MIT-licensed, a familiar API for Playwright users, resilience when websites change and smooth use from agents like Claude Code or Codex. Cons: it is a developer library, not a ready-made app; you need an AI model with your own key, and running it at scale usually means deploying on Browserbase's cloud infrastructure, which is a separate service. Performance figures are the project's own.