Overview

The Browserbase MCP Server gives AI models the ability to control real cloud browsers using Stagehand, an AI-powered browser automation framework. Unlike local Puppeteer or Playwright, Browserbase runs browsers in the cloud with stealth detection avoidance, making it ideal for web scraping, testing, and autonomous browsing.

Official Server:

Developed and maintained by Browserbase. Hosted MCP available at https://mcp.browserbase.com/mcp.

Key Features

🌐

Cloud Browsers

Headless browsers running in the cloud with no local resource consumption

🤖

Stagehand AI

AI-powered browser actions using natural language instructions via Stagehand

🕵️

Stealth Detection Avoidance

Built-in fingerprinting evasion and bot detection bypass

📊

Data Extraction

Extract structured data from web pages with natural language instructions

🎭

Session Persistence

Keep browser sessions alive across multiple interactions with Keep Alive mode

🎯

Element Observation

Observe and identify actionable elements on the page for interaction

Available Tools

Quick Reference

Tool	Purpose	Category
`start`	Create or reuse a Browserbase session	Session
`end`	Close the current Browserbase session	Session
`navigate`	Navigate to a URL	Navigation
`act`	Perform an action on the page	Interaction
`observe`	Observe actionable elements on the page	Observation
`extract`	Extract structured data from the page	Extraction

Detailed Usage

navigate▶

Navigate the browser to a specified URL. Must be called after starting a session.

{
  "url": "https://github.com/browserbase/mcp-server-browserbase"
}

act▶

Perform a browser action described in natural language. Powered by Stagehand AI.

{
  "action": "Click the star button on the repository page",
  "modelName": "google/gemini-2.5-flash-lite"
}

extract▶

Extract structured data from the current page using natural language instructions.

{
  "instruction": "Get the repository name, star count, and last commit date"
}

observe▶

Observe and identify actionable elements on the page, returning a list of candidates.

{
  "instruction": "Find all navigation links in the header"
}

Installation

{
  "mcpServers": {
    "browserbase": {
      "type": "http",
      "url": "https://mcp.browserbase.com/mcp"
    }
  }
}

Configuration Options

The self-hosted server accepts these command-line flags:

Flag	Description
`--proxies`	Enable Browserbase proxies for the session
`--verified`	Enable Verified Identity for bot detection avoidance
`--keepAlive`	Keep browser session alive across multiple interactions
`--modelName <model>`	Model for Stagehand (default: `google/gemini-2.5-flash-lite`)
`--modelApiKey <key>`	API key for custom model provider
`--browserWidth <px>`	Browser viewport width (default: 1024)
`--browserHeight <px>`	Browser viewport height (default: 768)

Docker Image:

Build the image first: git clone https://github.com/browserbase/mcp-server-browserbase.git && cd mcp-server-browserbase && docker build -t mcp-browserbase .

API Keys Required:

Sign up at browserbase.com for a Browserbase API key and project ID. A Gemini API key is required for Stagehand's AI-powered actions.

Security

API key authenticated — no direct system access to your machine
Browsers run in isolated cloud sandboxes, not on your local device
Stagehand actions are scoped to the current page only
Session isolation prevents cross-session data leakage
Hosted MCP uses HTTPS with OAuth authentication

Comparison: Browserbase vs Playwright/Puppeteer MCP

Feature	Browserbase	Playwright/Puppeteer MCP
Browser Location	Cloud (Browserbase infra)	Local machine
Bot Detection	Built-in stealth	Manual configuration
AI-Powered Actions	Stagehand (natural language)	Code-based selectors only
Session Persistence	Keep Alive mode	Per-command sessions
Infrastructure	Zero setup	Requires local browser install
Scaling	Cloud auto-scaling	Limited by local resources

Sources

Serper MCP Server

Serper MCP server provides Google Search via the Serper API — 13 specialised tools including web, images, news, scholar, patents, shopping, places, maps, and Lens search for LLM agents.

Flutter MCP Server

Flutter MCP servers enable AI models to interact with Flutter projects, providing capabilities for code analysis, formatting, testing, and documentation retrieval.

Figma MCP Server: Automate Design-to-Code Workflows

Connect Figma designs to VS Code and Cursor. Extract components, retrieve design tokens, and generate production-ready code with the Figma MCP server for seamless developer handoff.

Browserbase MCP Server