Best Puppeteer MCP in 2026: 5 Options for AI Browser Automation

Table of Contents

1. What Is a Puppeteer MCP Server?
2. Puppeteer MCP vs Playwright MCP
3. The Best Puppeteer MCP Servers
4. Puppeteer MCP Comparison
5. What You Can Actually Do With a Puppeteer MCP
6. Setup: Adding a Puppeteer MCP to Claude
7. Anti-Bot Detection: A Quick Reality Check
8. Performance and Resource Tips
9. Frequently Asked Questions

What Is a Puppeteer MCP Server?

A Puppeteer MCP server is a Model Context Protocol server that wraps Puppeteer (the Node library for controlling headless Chromium) and exposes its capabilities as tools an AI client can call. Once installed, your AI assistant can:

• Open a URL in a real Chromium browser
• Take a screenshot of a full page or a specific element
• Fill and submit forms
• Click links and follow them
• Extract text or HTML from selectors
• Run arbitrary JavaScript inside the page context
• Wait for selectors to appear (useful on JS-heavy sites)

The browser runs locally as a subprocess. When the AI client starts up, the Puppeteer MCP boots, registers its tools, and launches Chromium on demand. Once the task is done, the browser closes and resources are freed.

Browser automation MCPs are the second most powerful capability you can give an AI agent. Combined with the right model, they turn Claude into a true web agent that can research, scrape, fill forms, file complaints, book reservations, and run visual QA. Read the broader landscape in our best desktop MCP servers for 2026 guide.

Puppeteer MCP vs Playwright MCP

Both protocols give the AI a browser. The trade-offs in 2026:

• Puppeteer MCP is lighter, faster to install, and pinned to Chromium. The Chrome DevTools Protocol integration is tighter, which matters for advanced use cases like network interception and request modification.
• Playwright MCP supports Chromium, Firefox, and WebKit, has stronger auto-waiting (fewer flaky scripts), built-in video recording, and a better story for cross-browser test automation. It is now the default browser MCP shipped in many starter configs.

If you only need Chromium and want the smallest install, pick a Puppeteer MCP. If you need multi-browser support, robust auto-waits, or cross-browser CI tests, pick Playwright. Most production setups end up with both installed and use them for different jobs.

Browse Playwright options on Skiln at executeautomation Playwright or automatalabs Playwright.

The Best Puppeteer MCP Servers

1. Reference Puppeteer MCP

The original community-maintained Puppeteer MCP from the early days of the protocol. Solid tool surface: navigate, screenshot, click, fill, evaluate. No anti-bot evasion. Most stable for non-protected sites. Browse at Puppeteer MCP Server on Skiln.

Best for: Internal tools, dev environments, simple research workflows, scraping sites that do not run anti-bot checks.

Install: npx -y @modelcontextprotocol/server-puppeteer (the historical reference name; current builds may differ — check the Skiln listing).

2. Puppeteer Real Browser MCP

A variant built on puppeteer-real-browser that loads a stealth Chromium build to evade common bot detection. Bypasses Cloudflare Turnstile, FingerprintJS, and reCAPTCHA v3 on most sites. Slower to start than the reference MCP but worth it when you need it.

Best for: Scraping sites behind anti-bot protection, AI agents that need to act on consumer web properties (booking sites, ticket sales).

Install: Available via npm. Check Skiln for the current top-starred build.

3. MCP Scrape Copilot (Puppeteer)

A higher-level scraping MCP that wraps Puppeteer with a friendlier scrape API: pass a URL and a JSON schema, get back structured data. Handles pagination, infinite scroll, and shadow DOM out of the box. Useful when you do not want to write selectors. See the listing at MCP Scrape Copilot on Skiln.

Best for: Data extraction workflows, lead-generation pipelines, anyone who wants AI to do the selector-writing.

Install: See the README for the latest install command. Most variants ship as Node packages.

4. Puppeteer Testing MCP

A specialized Puppeteer MCP for QA and end-to-end testing. Exposes assertions, snapshot comparison, and integration with common test runners. Use it when you want Claude to write or run E2E tests against your own app. Browse the listing at Puppeteer Testing on Skiln.

Best for: QA teams, visual regression testing, AI agents that drive test suites.

Install: Available via npm. Check the project README.

5. When to Switch to Playwright MCP

If you find yourself fighting flakiness, needing Firefox or WebKit support, recording video for stakeholder review, or running parallel browser contexts at scale, switch to the Playwright MCP. It is the strictly more featureful tool in 2026, and most new browser MCP work happens on the Playwright side.

Best for: Anyone outgrowing the Puppeteer MCP toolkit. Multi-browser CI, production QA, agent loops that need stable auto-waits.

Install: npx -y @playwright/mcp@latest

Puppeteer MCP Comparison

What You Can Actually Do With a Puppeteer MCP

Six patterns that pay off the moment you install one:

• Take screenshots of pages mid-conversation. "Show me what the OG image looks like for skiln.co/blog/best-mcp-clients-2026." Claude navigates, screenshots, and inlines the image.
• Scrape competitor pricing. Hand Claude a list of URLs and a schema describing the data you want. Pair with the Postgres MCP to persist the rows.
• Fill out a long form with structured data. Walk Claude through a form once, then have it auto-fill the rest from a CSV. Useful for migrating between SaaS tools.
• Test your own site under different viewports. "Render https://skiln.co at 320px, 768px, and 1280px. Screenshot each. Tell me what breaks." Faster than firing up your own browser.
• Visual regression checks. Combined with Puppeteer Testing MCP, Claude can compare current screenshots against a saved baseline and flag diffs.
• Drive multi-step flows. Log in, navigate, click, submit. With the Sequential Thinking MCP at /mcp/sequential-thinking alongside, the loop stays coherent even across 20+ steps.

These patterns work in every major client. Read our MCP clients guide for client-specific browser support notes.

Setup: Adding a Puppeteer MCP to Claude

On Claude Desktop, edit your config:

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

Restart Claude Desktop. The first time the MCP runs, it will download a copy of Chromium (about 170MB). Subsequent runs reuse the cached binary.

For Claude Code:

claude mcp add puppeteer -- npx -y puppeteer-mcp-server

For Cursor and Windsurf, paste the same command into the MCP settings panel.

The Skiln Config Generator can bundle Puppeteer with the Git MCP, Postgres MCP, and Filesystem MCP into a single paste-ready config.

Anti-Bot Detection: A Quick Reality Check

Most modern consumer sites (Cloudflare-protected pages, ticket sales, airline booking, retail checkouts) actively detect headless browsers. Stock Puppeteer is detected by:

• Cloudflare Turnstile
• FingerprintJS Pro
• reCAPTCHA v3
• DataDome
• PerimeterX

The reference Puppeteer MCP will be blocked by all of them. Puppeteer Real Browser handles many, though not all. For the hardest cases, you need a dedicated stealth tool like CloakBrowser (we cover the install workflow in a separate skill).

If your scraping target is internal infrastructure or a dev site with no anti-bot layer, the reference MCP works fine. If it is a consumer property, expect to spend time tuning the stealth layer or to switch tools entirely.

Performance and Resource Tips

A running Chromium instance uses 200-400MB of RAM. Five tips to keep things smooth:

• Close the browser between tasks. The reference MCP does this by default. If you customize it, make sure cleanup runs on tool completion.
• Block image and font requests when you only need text. Cuts bandwidth and CPU in half on most sites.
• Disable JavaScript for scrape-only workflows where the content is server-rendered. Massive speedup.
• Cap concurrent pages. Puppeteer can run multiple pages in one browser. Two or three is fine. Ten will choke.
• Use the --headless flag explicitly. Some clients launch the MCP in headed mode by accident, which wastes resources.

Frequently Asked Questions

What is a Puppeteer MCP server?

A Puppeteer MCP server is a Model Context Protocol server that gives Claude (or another AI client) control of a headless Chromium browser through Puppeteer. The AI can navigate URLs, click elements, fill forms, take screenshots, scrape page content, and run arbitrary JavaScript on the page. Useful for web scraping, automation, visual QA, and AI agents that need to act on the live web.

Is Puppeteer MCP free?

Yes. Every Puppeteer MCP server we recommend is free and open source. You only pay for the resources the browser consumes on your machine (CPU and RAM, mostly) and for any AI tokens spent on the conversation.

Should I use Puppeteer MCP or Playwright MCP?

Playwright MCP is the more featureful option in 2026 and is now the default recommendation for most workflows. Puppeteer MCP still wins for two things: lighter resource footprint and tighter integration with the Chrome DevTools Protocol. If you only need Chromium and want the smallest possible install, pick Puppeteer. For multi-browser support, video recording, or auto-waiting, pick Playwright.

Can Puppeteer MCP bypass Cloudflare?

By default, no. Off-the-shelf Puppeteer is detected by most anti-bot systems (Cloudflare Turnstile, FingerprintJS, reCAPTCHA v3). For pages behind those protections, you need a stealth variant like Puppeteer Real Browser MCP or a higher-level tool like CloakBrowser. Even those are not guaranteed to work on every site.

Does Puppeteer MCP work with private intranets?

Yes. The browser runs locally on your machine, so any URL your machine can reach (including intranets, VPN-protected sites, and localhost dev servers) is accessible. Useful for in-house QA and tooling.

Will the AI take screenshots I don't want it to?

Only if you tell it to. Like every MCP, Puppeteer MCP exposes tools but the AI client decides when to call them. Claude Desktop prompts you before running browser tools by default. Claude Code can be configured with auto-approve rules — be conservative here.

How much RAM does Puppeteer MCP use?

Roughly 200-400MB per running browser instance, depending on the page complexity. The MCP shuts the browser down between calls in most implementations, so idle cost is minimal. For long-running agent loops that keep a browser open, plan on ~500MB of headroom.

Where can I see all browser-automation MCPs indexed on Skiln?

Visit /mcps and filter for 'browser' or 'scraping'. As of 2026, Skiln tracks 300+ browser automation MCPs covering Puppeteer, Playwright, Selenium, and emerging headless options like Stagehand and Browser-use.

Last updated: June 10, 2026 · Skiln tracks Puppeteer, Playwright, and Selenium MCP releases daily across 13 source registries.