Puppeteer MCP Server

@iatsiuk

リポジトリ

基本情報

カテゴリ

ブラウザ自動化

トランスポート

stdio

作者

iatsiuk

投稿者

Aleksei Iatsiuk

Puppeteer MCP Server について

MCP server for browser automation via Puppeteer. Executes arbitrary JavaScript with direct access to Puppeteer Browser instance. Single "execute" tool runs your code in a sandboxed Node.js VM with a browser global - full Puppeteer API access in one call.

#puppeteer #browser #automation #web-scraping #testing

Puppeteer MCP Server の設定

以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。

{
  "mcpServers": {
    "puppeteer": {
      "command": "npx",
      "args": [
        "pptr-mcp"
      ]
    }
  }
}

Puppeteer MCP Server のツール

Execute JS with Puppeteer 24.43.1. Globals: browser. Use 'return' for result. Use Puppeteer selectors: CSS, text/, aria/, xpath/, pierce/. Response includes screenshots (path, url) for each open tab.

概要

What is Puppeteer MCP?

Puppeteer MCP is an MCP server for browser automation that exposes a single execute tool, which runs arbitrary JavaScript code with direct access to the Puppeteer Browser instance. It is designed for trusted local development with LLM assistants like Claude Code and Cursor.

How to use Puppeteer MCP?

Install globally via npm install -g pptr-mcp, then add the server to your MCP client configuration (e.g., Claude Desktop) using npx pptr-mcp as the command. Optionally pass custom Chrome flags or set environment variables like CHROME_PATH, PUPPETEER_EXECUTABLE_PATH, and PPTR_MCP_TIMEOUT. Use the execute tool with a code parameter (JavaScript string) and optional persistent boolean (default true) to reuse the browser session.

Key features of Puppeteer MCP

Single execute tool runs arbitrary Puppeteer JavaScript code.
Full Puppeteer API access within one round-trip call.
Persistent browser session across calls (configurable).
Supports custom Chrome flags (window size, proxy, etc.).
Works as a Claude Code plugin via /plugin add.

Use cases of Puppeteer MCP

Automating complex multi-step browser workflows in one call.
Taking screenshots, scraping content, or filling forms with custom logic.
Testing web applications by executing Puppeteer scripts from an LLM.
Debugging and interacting with pages that require custom navigation or state.

よくある質問

How is Puppeteer MCP different from other browser MCP servers?

Most browser MCP servers expose fixed tools (navigate, click, screenshot) requiring multiple round-trips. Puppeteer MCP uses a single `execute` tool that runs your own Puppeteer code, handling complex workflows in one call.

What are the security considerations?

This server is for trusted local development only. The Node.js VM isolates code for convenience, not security. Executed code has full browser control, Chrome runs with `--no-sandbox` for Docker compatibility, and persistent sessions preserve state across calls. Do not use in multi-tenant or untrusted environments.

What dependencies are required?

Puppeteer and a Chrome browser. The server will use the default Puppeteer-bundled Chromium unless `CHROME_PATH` or `PUPPETEER_EXECUTABLE_PATH` environment variables are set.

Can I reuse browser sessions across calls?

Yes. The `persistent` parameter defaults to `true`, preserving cookies and state across calls. Set `persistent: false` for session isolation.

What transport and authentication does it use?

The server communicates over stdio via the MCP protocol, as configured in the MCP client. No authentication is built in; it is intended for local use with trusted clients.

「ブラウザ自動化」の他のコンテンツ

TinyFish

tinyfish

Give your AI agent the live web. One MCP server for Search, Fetch, Browser, and Agent: browse real sites, extract structured data, fill forms, and run multi-step workflows. Search and Fetch are free.

Open Browser Use

iFurySt

🔮 Platform-neutral Browser Use for AI agents: real Chrome automation with a CLI + SDKs, no lock-in, dead simple.

371

Fastcrw

Fast, lightweight Firecrawl/Tavily alternative in Rust. Web scraper, crawler & search API with MCP server for AI agents. Drop-in Firecrawl-compatible API (/scrape, /crawl, /search). 2.3x faster than Tavily, 1.5x faster than Firecrawl in 1K-URL benchmarks. 6 MB RAM, single binary.

286