Browserbase MCP Server

What It Does

Cloud-hosted browser automation that runs in Browserbase’s infrastructure instead of your local machine. The Stagehand AI SDK powers high-level browser interactions — act, extract, observe — using an LLM internally to understand page structure. Features include stealth mode, proxy support, anti-bot bypass, session persistence across requests, and Shadow DOM/iframe support.

This is for complex interactive automation where local Playwright or Puppeteer would be blocked or insufficient. Simple page fetching or scraping doesn’t need this.

What It Does Well

• Stealth mode and anti-bot bypass handle sites that block automation. If your target pages detect headless browsers, require CAPTCHA solving, or use aggressive bot detection, Browserbase’s cloud infrastructure handles the evasion. Local Playwright and Puppeteer get blocked on these sites.
• Session persistence lets Claude maintain state across interactions. Browser contexts persist between MCP calls, so Claude can log into a site in one request and continue interacting as a logged-in user in subsequent requests. Local MCP browser servers lose state when the tool call ends.
• Stagehand’s AI-driven page understanding reduces brittle selectors. Instead of writing CSS selectors that break when a site redesigns, Stagehand uses an LLM to understand page structure semantically. “Click the login button” works even when the button’s class name changes.

What It Doesn’t Do Well

• Nested LLM dependency adds cost and complexity. Stagehand uses its own LLM (Gemini by default) internally to understand pages. You’re paying for Claude’s API calls plus a second model’s API calls plus Browserbase’s service fee. Three cost layers for browser automation.
• Overkill for most scraping tasks. If you just need to extract text from a page, Firecrawl or Fetch MCP do it faster and cheaper. Browserbase’s value is in interactive automation with anti-bot requirements — a minority of MCP browser use cases.

Setup Notes

Three credentials needed: Browserbase API key, project ID, and an LLM API key for Stagehand. The LLM defaults to Gemini but supports Anthropic and OpenAI. Setup is advanced — rated 3/3 — because of the multi-credential requirement and the conceptual overhead of understanding what Stagehand does with its internal LLM.

Docker is also supported if you prefer containerized setup.

Config

{
  "mcpServers": {
    "browserbase": {
      "command": "npx",
      "args": ["-y", "@browserbasehq/mcp-server-browserbase"],
      "env": {
        "BROWSERBASE_API_KEY": "your_browserbase_key",
        "BROWSERBASE_PROJECT_ID": "your_project_id"
      }
    }
  }
}

Tested With

• Claude Desktop on Windows 11

{
  "mcpServers": {
    "browserbase-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@browserbasehq/mcp-server-browserbase"
      ],
      "env": {
        "BROWSERBASE_API_KEY": "your_browserbase_key",
        "BROWSERBASE_PROJECT_ID": "your_project_id"
      }
    }
  }
}