Potencialize seu agente
O comando do próprio publicador. Não é um endpoint do PlugMCP.
Host a copy with our keys and logs.
npx -y [email protected]O Claude Code não tem link de instalação para um comando local. Copie o comando abaixo e execute-o em um terminal.
claude mcp add puppeteer-mcp-claude -- npx -y puppeteer-mcp-claude@0.2.2Execute em um terminal.
{ "mcpServers": { "puppeteer-mcp-claude": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
URL remota ou um comando na sua máquina? Como as duas instalações diferem
Rodar no PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
AbrirA Model Context Protocol server that gives Claude Code (and any other MCP-aware client) a real browser via Puppeteer — navigate pages, click and type, run JavaScript, capture screenshots, manage cookies, intercept requests.
macOS / Linux
curl -fsSL https://raw.githubusercontent.com/jaenster/puppeteer-mcp-claude/main/install.sh | bash
Windows (PowerShell)
iwr -useb https://raw.githubusercontent.com/jaenster/puppeteer-mcp-claude/main/install.ps1 | iex
Both scripts verify Node ≥ 18, install the package globally via npm, and register it with Claude Code at user scope. Override scope with SCOPE=project (bash) or $env:SCOPE='project' (PowerShell).
Manual — if you'd rather not run a remote script:
npm install -g puppeteer-mcp-claude
claude mcp add puppeteer-mcp-claude -- npx -y puppeteer-mcp-claude serve
Then restart Claude Code and ask: "Take a screenshot of example.com".
You don't have to call puppeteer_launch first — the browser auto-launches with defaults the moment any other tool runs. Page ids default to "default", so single-tab flows can omit pageId entirely.
Every tool returns its result in two parallel forms:
content[0].text — TOON (Token-Oriented Object Notation), a compact, schema-aware JSON alternative. Good for hosts that pipe the text body straight into the model's context.structuredContent — the same data as a typed JSON object, for MCP clients that prefer machine-readable output.Which one your host uses is up to the host — both are MCP-spec-compliant. Typed clients that want the object without depending on structuredContent can import { decode } from '@toon-format/toon' and parse content[0].text.
How an LLM would typically drive this server — each example shows the prompt you'd give Claude and the tool sequence it produces.
"Take a screenshot of news.ycombinator.com."
puppeteer_navigate { url: "https://news.ycombinator.com" }
puppeteer_screenshot { fullPage: true }
Returns the PNG as an inline image block (Claude sees it directly), plus structured { bytes, path: null, fullPage: true }.
"Give me the titles of the top 10 stories on Hacker News as a markdown list."
puppeteer_navigate { url: "https://news.ycombinator.com" }
puppeteer_evaluate { script: "[...document.querySelectorAll('.titleline > a')].slice(0,10).map(a => a.textContent)" }
Claude formats the resulting array into the requested markdown list.
"On example.com, fill the search box with 'TOON' and submit."
puppeteer_navigate { url: "https://example.com" }
puppeteer_type { selector: "input[name=q]", text: "TOON" }
puppeteer_click { selector: "button[type=submit]" }
puppeteer_wait_for_selector { selector: ".results" }
puppeteer_get_text { selector: ".results h1" }
"Connect to my running Chrome and take a screenshot of the GitHub dashboard."
# In one terminal
puppeteer-mcp-claude chrome 9222
# Sign in to GitHub once in the launched Chrome window.
puppeteer_launch { browserWSEndpoint: "ws://localhost:9222" }
puppeteer_navigate { url: "https://github.com" }
puppeteer_screenshot {}
Avoids re-doing login flows in every session.
"Scrape the article text from as quickly as possible."
puppeteer_set_request_interception { enable: true, blockResources: ["image","media","font","stylesheet"] }
puppeteer_navigate { url: "<url>" }
puppeteer_get_text { selector: "article" }
puppeteer-mcp-claude install [--scope user|project|local] Register with Claude Code
puppeteer-mcp-claude uninstall [--scope ...] Remove from Claude Code
puppeteer-mcp-claude status Show "claude mcp list"
puppeteer-mcp-claude serve Run the MCP server on stdio
puppeteer-mcp-claude chrome [port] [userDataDir] Launch Chrome with remote debugging
puppeteer-mcp-claude help Show help
pnpm install
pnpm build
pnpm test # node:test via tsx, ~165 tests
pnpm dev # run the server directly from src/
MIT
Browser automation MCP server for Claude, powered by Puppeteer.
{ "mcpServers": { "puppeteer-mcp-claude": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}Ferramentas
Ainda não há lista de ferramentas. Servidores remotos recebem uma após a primeira verificação de saúde.
Perguntas frequentes
Copie nesta página a configuração stdio para Cursor, Claude Desktop, Claude Code ou VS Code. Ela usa a instalação do publicador, não um proxy do PlugMCP.
Este registro não lista nenhuma variável de ambiente obrigatória.
Puppeteer MCP Claude roda localmente via stdio.
Última verificação: 2026-11-16. Browser automation MCP server for Claude, powered by Puppeteer.