Cloud browsers for AI agents — automation sessions that run in Browserbase's fleet, not on your machine.
Browserbase answers the question every local browser automation setup eventually raises: where should these browsers actually run? Its MCP server hands your assistant a browser session hosted in Browserbase's cloud fleet — with stealth options, proxies, and session recordings — while your own machine does nothing but relay commands.
The practical wins are real. Local browser automation ties up your laptop, breaks in CI without a display stack, and gets blocked by sites that fingerprint obvious headless Chromium. Cloud sessions sidestep all three: they scale horizontally, come with residential proxy options for sites that geo-gate or rate-limit, and every session leaves a replay you can watch to debug what the agent actually did.
Under the hood the server pairs with Stagehand, Browserbase's AI-native automation layer, so the model can act on pages using natural-language instructions ("click the pricing toggle") rather than brittle selectors. For extraction work you describe the data you want and get structured output back.
It's a paid service with a free tier, and that's the honest trade-off versus the Playwright server: you're exchanging money for reliability at scale, anti-bot resilience and zero local setup. For an occasional scrape, run Playwright locally. For an agent that browses hundreds of pages a day as part of a product, cloud sessions are usually the difference between a demo and something you can ship.
browserbase_session_close rather than waiting for the idle timeout. An agent that forgets is an agent that runs a browser all night.browserbase_act when the page is unfamiliar and plain selectors when it isn't; each natural-language action costs an extra model round trip through Stagehand.Compare MCP servers by category, or browse hosted MCP servers if you prefer an endpoint over a local process.
Products whose agents visit many pages per day, where local browsers would be a bottleneck or a liability.
Stealth mode and proxy support get through defences that instantly flag vanilla headless Chromium.
Every session is recorded, so you can replay exactly what the agent clicked when something goes wrong.
Check the linked publisher documentation and your connected client's tool list. Names and availability can vary with the server version and account permissions.
| Tool | What it does |
|---|---|
| browserbase_session_create | Start a fresh cloud browser session |
| browserbase_navigate | Navigate the session to a URL |
| browserbase_act | Perform an action described in natural language ('click Sign in') |
| browserbase_extract | Extract structured data from the current page |
| browserbase_screenshot | Capture a screenshot of the cloud session |
| browserbase_session_close | End the session and stop billing for it |
{
"mcpServers": {
"browserbase": {
"command": "npx",
"args": ["-y", "@browserbasehq/mcp"],
"env": {
"BROWSERBASE_API_KEY": "your-api-key",
"BROWSERBASE_PROJECT_ID": "your-project-id"
}
}
}
}A Browserbase account (free tier available). Node.js 18+ for the local connector.
| Variable | Description | Required |
|---|---|---|
| BROWSERBASE_API_KEY | API key from the Browserbase dashboarde.g. bb_live_... | Yes |
| BROWSERBASE_PROJECT_ID | Project ID the sessions bill against | Yes |
Use Microsoft’s Playwright MCP server to navigate pages, inspect accessibility snapshots, fill forms and capture browser screenshots.
The original Chromium automation reference server — simple, screenshot-driven browser control.
Yet another WebUI for Nginx
I use your Instagram Web browser session (Integrations → Instagram Web) via instagrapi, not Meta / Graph API
Self-hosted Chrome for AI agents, with a live noVNC view a human can take over.
Find the right ToolsMonk tool for a task: 255 free browser-based tools, read-only lookups.