Browser Automation

Stagehand

A browser agent SDK that extracts data and interacts with any site through act, observe, and extract calls.

Quick decision

Best for: Developers who want explicit, readable agent steps with self-healing actions and structured extraction, in TypeScript, Python, or Go.
Not ideal for: Fully autonomous, open-ended task execution. Stagehand expects you to direct the steps, which is a strength for control and a limit for hands-off work.
Pricing model: The SDK is open source under MIT. Local runs need Chrome installed. Browserbase cloud browsers and the hosted MCP server are a separate hosted service. See Browserbase for current hosted pricing.
Deployment: Self-hosted library. Hosted cloud (optional). Hosted MCP server (optional).
Authentication: Local runs need a model API key (for example OPENAI_API_KEY). Browserbase runs use a BROWSERBASE_API_KEY. Credentials stay in your environment, not in page code examples.
Review state: Source verified. Facts checked 2026-10-03. Not locally tested by AgentsUse.

What it does

Stagehand is an SDK for browser agents from the Browserbase team. It pairs Playwright-style page methods with AI-assisted calls: act carries out an instruction, observe finds elements and returns real selectors, and extract returns schema-validated data.

The SDK is published for TypeScript, Python, and Go. It can drive a local browser with a persistent profile, so a signed-in session carries over to the next run, or run on Browserbase cloud browsers.

A hosted Browserbase MCP server exposes navigate, act, observe, and extract to MCP clients without a local install. The same project also documents search and fetch helpers that skip the browser when full rendering is not needed.

Verified capabilities

Browser control

Natural-language actions Source verified
act carries out an instruction such as clicking a sign-in button and self-heals when the site changes.
Element discovery Source verified
observe returns real selectors for an instruction, so credentials and values need not pass through the model.
Persistent local sessions Source verified
A local user data directory keeps cookies, so the next run starts signed in.

Data access

Schema-validated extraction Source verified
extract returns data validated against a schema you define (Zod, Pydantic, or Go types).

Setup

Three language SDKs Source verified
TypeScript, Python, and Go packages with the same core model.

Agent integration

Hosted MCP server Source verified
Browserbase hosts an MCP endpoint with navigate, act, observe, and extract tools.

Quick start

Install Stagehand

From the official README. Local runs need Chrome installed. Python and Go installs use the same package names shown in the README.

pnpm add @browserbasehq/stagehand zod
# Python: pip install stagehand

Environment variables (names only, use your own values)

OPENAI_API_KEY or another model key
BROWSERBASE_API_KEY (only for Browserbase cloud)

Source: https://github.com/browserbase/stagehand. Examples use placeholders only. Never paste a real key into a profile, config file you share, or a ticket.

MCP support: Yes, hosted. Browserbase documents a hosted MCP server at mcp.browserbase.com with navigate, act, observe, and extract tools. AgentsUse has not locally tested it.

Works with

MCP

Hosted Browserbase MCP server (HTTP transport) documented by the project.

Packages

npm: @browserbasehq/stagehand. PyPI: stagehand. Go module published by the project.

Browsers

Local Chrome or Browserbase cloud browsers.

Deployment

Self-hosted SDK, optional Browserbase cloud, optional hosted MCP.

Only sourced support is listed. A missing framework means AgentsUse has not verified it yet, not that it cannot work.

Health and maintenance

GitHub stars: 25,529 (checked 2026-10-03)
GitHub forks: 1,752 (checked 2026-10-03)
Package: npm: @browserbasehq/stagehand
License: Open source, MIT
Maintainer: Browserbase

Stars and forks from the GitHub repository page. No package version is shown because AgentsUse did not re-check the npm version field in this pass. Maintenance signals only, not a quality rating.

Pricing and license

The SDK is open source under MIT. Local runs need Chrome installed. Browserbase cloud browsers and the hosted MCP server are a separate hosted service. See Browserbase for current hosted pricing.

Official pricing or docs →

Limitations and safety

Act and extract depend on the model you connect. Budget model calls and review extracted data before acting on it.
The fastest cloud path is a hosted service. Local runs avoid that dependency but need Chrome and your own machine.
No independently checked benchmark is published by AgentsUse for this tool yet. Speed claims on the project site are vendor claims, not AgentsUse results.

Browser and data tools can read pages, fill forms, and download files. Start with a test account or read-only access, keep credentials in environment variables, and review agent actions before connecting anything that can spend money, send messages, or delete data.

Alternatives to Stagehand

Playwright

Choose Playwright when you want mature, deterministic automation and testing without a model in the loop.

Tradeoff: Selectors and steps are yours to maintain when sites change.

Browser Use

Choose Browser Use when you want a Python agent that plans a whole task from a description.

Tradeoff: Less step-by-step control, and outcomes depend more heavily on the model.

Puppeteer

Choose Puppeteer when you want a long-established JavaScript API for Chrome with no AI layer.

Tradeoff: No self-healing actions or schema extraction built in.

Related tools

Playwright

Browser Automation

Browser Use

Browser Automation

Puppeteer

Browser Automation

Common questions

What does Stagehand do for an AI agent?

A browser agent SDK that extracts data and interacts with any site through act, observe, and extract calls.

Is Stagehand open source?

Yes. This profile records the license as MIT from the official repository.

Does Stagehand support MCP?

Yes, hosted. Browserbase documents a hosted MCP server at mcp.browserbase.com with navigate, act, observe, and extract tools. AgentsUse has not locally tested it.

Sources and freshness

Last checked 2026-10-03. Verification label: source verified, which means public claims trace to the sources above. It does not mean AgentsUse ran the tool. Spotted an error? Send a correction. Back to the tools directory.