Best Puppeteer MCP Servers in 2026: 5 Tools for AI Browser Automation

Table of Contents

1. What Is a Puppeteer MCP Server?
2. Puppeteer MCP vs Playwright MCP
3. How We Ranked These
4. The Best Puppeteer MCP Servers
5. Comparison Table
6. Real Use Cases
7. Setup: Puppeteer MCP in 60 Seconds
8. Headed vs Headless Mode
9. Performance & Resource Use
10. Frequently Asked Questions

What Is a Puppeteer MCP Server?

A Puppeteer MCP server is a Model Context Protocol bridge between an AI client and a real Chromium browser controlled by Google's Puppeteer library. When you install one, your AI assistant gains the ability to:

• Navigate to any URL and wait for it to load
• Click buttons, fill forms, select dropdowns
• Take screenshots (full page or selector-scoped)
• Scrape rendered HTML (post-JavaScript execution)
• Evaluate arbitrary JavaScript in the page context
• Intercept network requests and capture responses
• Manage cookies, localStorage, and authenticated sessions

Puppeteer launched in 2017 as the canonical Chromium automation library and remained dominant through 2024. Today, Playwright is the bigger ecosystem, but Puppeteer still wins for Chrome DevTools Protocol access, CDP-specific features (heap snapshots, accessibility tree dumps, coverage reports), and legacy compatibility with existing scrape scripts.

Puppeteer MCP vs Playwright MCP

A common question. Quick decision tree:

• Greenfield project, no existing scripts? Use Playwright MCP. Cross-browser support (Chrome + Firefox + WebKit), better selector ergonomics, first-party Microsoft maintenance.
• Existing Puppeteer scripts you want the AI to reuse? Use Puppeteer MCP. The tool surface maps 1:1 to your existing code.
• Need CDP / Chrome DevTools access? Use Puppeteer MCP. Playwright wraps CDP partially — Puppeteer exposes it natively.
• Need stealth / bypass anti-bot? Either works, but you'll want a stealth plugin or stealth fork in both cases.

For most teams in 2026, Playwright is the default and Puppeteer is the fallback. We still recommend keeping Puppeteer in your toolbox — the use cases where it wins are narrow but real.

How We Ranked These

Six signals:

1. API coverage — how many Puppeteer methods are surfaced as MCP tools?
2. Reliability — do timeouts, retries, and error messages behave sensibly?
3. Configuration depth — can you swap viewport size, user agent, proxy, headless mode?
4. Stealth support — does it ship with or document anti-detection options?
5. Resource cleanup — does it close browser instances after each session?
6. Community maintenance — stars, recent commits, real-world production use.

The Best Puppeteer MCP Servers

1. merajmehrabi Puppeteer MCP

The community favorite. Maintained actively. Full API coverage — navigate, click, fill, screenshot, evaluate, intercept. Sensible defaults (30-second navigation timeout, headless by default), with overrides for every option. Surfaces over 20 tools to the AI client.

Best for: General-purpose browser automation, scraping, form filling, screenshot capture.

Install: Listed on Skiln — check the project README for the current install command.

2. Configurable Puppeteer MCP

PulseMCP-listed Puppeteer variant with a heavy focus on configurability. Exposes runtime settings (viewport, user agent, proxy, locale) as MCP tool parameters rather than env variables. Useful when one AI workflow needs to switch contexts between automation runs.

Best for: Multi-tenant scraping, A/B-style automation tests, workflows where every run needs different browser settings.

Install: See the PulseMCP listing for the exact install command and configuration recipe.

3. Smithery Puppeteer Browser

Smithery's hosted-and-local Puppeteer MCP. Built for the Smithery installation flow (one-click install via the Smithery CLI). Slightly fewer tools than merajmehrabi's MCP, but the install ergonomics are unmatched — especially for users already using Smithery to manage their MCP stack.

Best for: Smithery users, teams that prefer one-click install over hand-crafted configs.

Install: npx -y @smithery/cli install puppeteer (assuming Smithery CLI installed).

4. Puppeteer Automation Skill

A Claude-skill-formatted Puppeteer integration listed in the SkillsDirectory. Not strictly an MCP server — it's a Claude skill that bundles Puppeteer with workflow recipes (login flows, screenshot pipelines, scraping templates). Useful if you'd rather have a skill than configure an MCP server.

Best for: Users on Claude Pro/Max who prefer skills to MCPs, structured automation recipes.

Install: Install via Skiln's skill installer or copy directly into your .claude/skills/ folder.

5. LobeHub Puppeteer MCP Guide

A LobeHub-curated Puppeteer MCP packaged with a getting-started guide and tested recipe configs. Best entry point for someone who's never wired up a browser-automation MCP before. Less powerful than the merajmehrabi MCP but far easier to follow if you're learning.

Best for: First-time Puppeteer MCP users, tutorial-driven workflows, learning the browser-automation toolset.

Install: Check the LobeHub page on Skiln for the current install pattern.

Comparison Table

Real Use Cases

Where Puppeteer MCP earns its keep:

• Login + scrape flows. Sites behind auth that need a real browser to render. Claude logs in, navigates, scrapes the rendered HTML, and returns structured data.
• Visual QA. Take screenshots of staging vs production for the same URL, compare them with a vision-capable model, flag regressions.
• Form automation. Fill out long applications, multi-step checkouts, government forms. Faster than HTTP-only scraping when forms validate client-side.
• PDF generation from web pages. Render an HTML report in headless Chrome and capture it as a PDF. Perfect for invoice generation, report exports.
• End-to-end test prototyping. Have Claude write Puppeteer-style integration tests by recording your manual click-through, then converting to a script.
• Web archiving. Capture full-page snapshots of dynamic pages for compliance, archival, or training-data collection.

Setup: Puppeteer MCP in 60 Seconds

Most Puppeteer MCPs install via npm. Example with the merajmehrabi MCP:

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

Drop into your Claude Desktop or Claude Code config. Restart the client. The puppeteer tool surfaces with ~20 sub-tools (navigate, click, fill, screenshot, evaluate, etc.).

For Claude Code, the one-liner:

claude mcp add puppeteer -- npx -y @merajmehrabi/puppeteer-mcp-server

First navigation triggers Puppeteer's Chromium download (~170 MB). Subsequent runs reuse the cached binary.

The Skiln Config Generator bundles Puppeteer with related MCPs (Playwright, Firecrawl, Browserbase) into one config block.

Headed vs Headless Mode

Default: headless (no visible window). Good for production, bad for debugging.

Pass HEADLESS=false (or the equivalent flag for your chosen MCP) when you want to watch the browser drive itself. Useful for:

• Debugging selector mismatches
• Testing visual workflows
• Demo-ing automation to stakeholders
• Sites that fingerprint headless browsers (LinkedIn, Twitter)

For sites with aggressive bot detection, headed mode alone isn't enough. Layer in puppeteer-extra-plugin-stealth or switch to a stealth fork. The relevant configuration is documented in every Puppeteer MCP's README on Skiln.

Performance & Resource Use

Each running Puppeteer browser is ~150-300 MB of RAM and one Chromium process. Practical limits:

• Local laptop: 3-5 concurrent browsers before you feel it.
• Mid-tier VPS (4 GB RAM): 5-10 concurrent browsers.
• High-volume scraping: Use a remote browser pool (Browserless, Browserbase) accessed through their MCPs instead of running browsers locally.

The MCP itself adds minimal overhead — almost all the cost is Chromium. Cache browser instances when possible (keep-alive mode in the merajmehrabi MCP) to avoid the startup cost on every tool call.

Cross-reference with the desktop MCP guide for browser-automation alternatives like the official Playwright MCP and Firecrawl MCP, which we benchmarked side-by-side.

Frequently Asked Questions

What is the Puppeteer MCP server?

The Puppeteer MCP server is a Model Context Protocol server that gives an AI client control over a real Chromium browser via Google's Puppeteer library. It exposes tools to navigate URLs, click elements, fill forms, take screenshots, evaluate JavaScript, and capture network traffic — all through MCP. Most workflows can be built with the merajmehrabi or Configurable Puppeteer MCPs as the starting point.

Puppeteer MCP vs Playwright MCP — which should I use?

Playwright MCP is the modern recommendation. It supports Chromium, Firefox, and WebKit, has better selector ergonomics, and ships with first-party MCP support from Microsoft. Puppeteer MCP is still a strong choice if you're tied to Chrome-specific APIs (CDP access, Chrome DevTools Protocol), or if your existing scripts are Puppeteer-based. See our Playwright MCP vs Puppeteer MCP comparison for the deep dive.

Does Puppeteer MCP run headless or with a visible browser?

Both are supported. By default most Puppeteer MCPs run headless (no visible window). Pass HEADLESS=false (or the equivalent flag for the specific MCP you chose) to get a real Chrome window. Headed mode is useful when debugging selectors or when a site blocks headless browsers via fingerprinting.

Can Puppeteer MCP take screenshots?

Yes — every major Puppeteer MCP exposes a screenshot tool. Most return PNG bytes inline so the AI can analyze the image directly. The screenshot tool typically accepts width, height, full-page flag, and selector-scoped screenshots, matching the Puppeteer API closely.

Can Puppeteer MCP bypass Cloudflare or reCAPTCHA?

Not out of the box. Stock Puppeteer is detected by Cloudflare Turnstile, FingerprintJS, and reCAPTCHA v3 in most cases. If you need stealth, layer in a stealth plugin (puppeteer-extra-plugin-stealth) or switch to a stealth browser like Camoufox. We documented the bypass options in our scraping guides on Skiln.

Is Puppeteer MCP free?

Yes. Every Puppeteer MCP we ranked is open source and free. The only cost is the underlying compute — each Puppeteer instance runs a Chromium process, which uses ~150-300 MB of RAM. For high-volume workflows, consider a remote browser pool (Browserless, Browserbase) accessed through their dedicated MCPs instead of running browsers locally.

Will Puppeteer MCP work on Windows / Mac / Linux?

Yes — all three. Puppeteer downloads a pinned Chromium binary on first run, so you don't need a system-wide Chrome install. On Linux you may need to install fonts and a few system libraries (libnss3, libatk1.0-0). The full dep list is in the Puppeteer README on Skiln's listing page.

Where can I see all Puppeteer-related MCPs on Skiln?

Visit /mcps and search for puppeteer. Skiln indexes 50+ Puppeteer-flavored MCPs across the major forks, scraping helpers, and specialized variants like form-fillers and visual-regression tools.

Last updated: May 27, 2026 · Skiln tracks every browser-automation MCP across 13 source registries every day.