browserbase

Official
Local
3.1k
Signed
GitHub Repo

Overview

The browserbase MCP server is a Model Context Protocol (MCP) server that gives AI assistants full, programmable control of real web browsers running in the cloud. Powered by Browserbase, it enables AI-driven workflows to load pages, execute JavaScript, interact with dynamic UIs, and capture screenshots or HTML — making it possible to automate and reason over modern, JavaScript-heavy websites that traditional HTTP-based tools can’t handle.

This server is ideal for web automation, end-to-end testing, live research, and data extraction workflows that require a real browser environment.

Transport

stdio

Tools

  • createSession
  • listSessions
  • closeSession
  • navigateWithSession
  • actWithSession
  • extractWithSession
  • observeWithSession
  • getUrlWithSession
  • getAllUrlsWithSession
  • closeAllSessions
  • navigate
  • act
  • extract
  • observe
  • screenshot
  • getUrl

Key Capabilities

  • Real browser automation — Control headless or full browsers that fully execute JavaScript and render dynamic content.
  • Modern web compatibility — Interact with SPAs, authenticated flows, and sites that rely heavily on client-side rendering.
  • Scriptable interactions — Click buttons, fill forms, scroll pages, and run custom JavaScript logic.
  • Visual inspection — Capture screenshots for validation, debugging, or visual analysis inside AI workflows.
  • Session isolation — Each browser session runs independently for safe, repeatable automation.

How It Works

The browserbase MCP server runs as an MCP service that connects AI clients to Browserbase’s managed browser infrastructure. When an agent invokes a tool, the server provisions a remote browser session and exposes it to the AI client through MCP.

Subsequent tool calls are translated into browser automation commands that run in real time within the session. The server captures results (HTML, screenshots, or execution output) and returns them in structured form over the MCP protocol.

Authentication is handled via a Browserbase API key configured in the environment, and all browser lifecycle management (startup, teardown, isolation) is handled by the server. This allows AI assistants to safely and reliably interact with the live web — performing tasks like form submission, data extraction from dynamic pages, or end-to-end workflow automation — without embedding browser automation logic directly into the client.