Browser Use Mcp Server
@bytedance
Browser automation capabilities using Puppeteer. This server enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models.
概要
What is Browser Use MCP Server?
Browser Use MCP Server is a Model Context Protocol (MCP) server that provides browser automation capabilities using Puppeteer. It enables LLMs to interact with web pages through structured accessibility snapshots, bypassing the need for screenshots or visually-tuned models.
How to use Browser Use MCP Server?
Install by adding a JSON configuration to your MCP client (VS Code, Cursor, Windsurf, Claude Desktop, or any MCP client) using the command npx @agent-infra/mcp-server-browser@latest. The server supports local (stdio) mode by default, remote mode via SSE or Streamable HTTP by specifying a --port argument, and in-memory calls for JavaScript/TypeScript projects. Key configuration arguments include --browser, --cdp-endpoint, --headless, --proxy-server, --user-agent, and --viewport-size.
Key features of Browser Use MCP Server
- Fast and lightweight using Puppeteer's label index and accessibility DOM tree.
- LLM-optimized: operates on structured data, no vision models needed.
- Cross-platform and extensible with support for remote and local browsers.
- Works with any MCP client: VS Code, Cursor, Windsurf, Claude Desktop.
- Supports multiple transport modes: stdio, SSE, and Streamable HTTP.
- Offers in-memory integration for JavaScript/TypeScript MCP clients.
Use cases of Browser Use MCP Server
- LLM agents navigating and interacting with web pages via structured accessibility snapshots.
- Automating browser tasks without requiring visual or screenshot-based models.
- Extracting structured data from web pages in a token-efficient manner.
- Running browser automation remotely over SSE or Streamable HTTP.
FAQ from Browser Use MCP Server
What makes this different from screenshot-based browser automation?
It uses Puppeteer's accessibility DOM tree and label index instead of pixel-based input, requiring no vision models and consuming less context tokens.
What are the runtime requirements?
Node.js 18 or newer is required, along with any MCP client such as VS Code, Cursor, Windsurf, or Claude Desktop.
How does the server connect to a browser?
It can launch a local browser (default), connect via a CDP endpoint (--cdp-endpoint), a WebSocket endpoint (--ws-endpoint), or use a custom browser executable path (--executable-path).
What transport protocols are supported?
The server supports the standard MCP stdio transport (local), SSE transport, and Streamable HTTP transport for remote connections.
Can I use a proxy with Browser Use MCP Server?
Yes, you can specify a proxy server with --proxy-server and bypass domains with --proxy-bypass.