1. Conduid
  2. AI
  3. Website
MCP server · AI

Website

Quickly screenshots webpages and converts to an LLM friendly size

Unclaimed MIT last commit 11 months ago ai
56Fair

Scored 5 months ago · breakdown

About Website

Website is an MCP server published by chongzixuan-ai in the AI category: quickly screenshots webpages and converts to an LLM friendly size.

The repository has 26 stars and 1 forks, with the last commit 11 months ago. Six months or more without a commit doesn't mean the server is broken, but check the open issues before depending on it in production.

Install

From source
git clone https://github.com/chongzixuan-ai/mcp-server-website

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Website

Powered by Claude · Grounded in docs

I know everything about Website. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

Static checks derived from the repository 5 days ago. Heuristics on the source, not an audit.

  • ✓README presentThe repository ships documentation.
  • ✓License declaredUsable under a stated license.
  • ✓Tests presentA test suite exists in the repository.
  • ✓Dependencies pinnedA lockfile pins dependency versions.
  • ✓No dynamic code executionSource scan found no eval, dynamic Function, or shell execution.
  • !Scoped permissionsDoesn't declare a permission scope. Assume it can do anything its process can.

README

@just-every/mcp-screenshot-website-fast

Fast, efficient screenshot capture of web pages - optimized for CLI coding tools. Automatically tiles full pages into 1072x1072 chunks for optimal processing.

npm version GitHub Actions

Overview

Built specifically for AI vision workflows, this tool captures high-quality screenshots with automatic resolution limiting and tiling for optimal processing by Claude Vision API and other AI models. It ensures screenshots are perfectly sized at 1072x1072 pixels (1.15 megapixels) for maximum compatibility.

Features

  • 📸 Fast screenshot capture using Puppeteer headless browser
  • 🎯 Claude Vision optimized with automatic resolution limiting (1072x1072 for optimal 1.15 megapixels)
  • 🔲 Automatic tiling - Full pages are automatically split into 1072x1072 tiles
  • 🎬 Screencast capture - Record series of screenshots over time with configurable intervals
  • 🔄 Always fresh content - No caching ensures up-to-date screenshots
  • 📱 Configurable viewports for responsive testing
  • ⏱️ Wait strategies for dynamic content (networkidle, custom delays)
  • 📄 Full page capture by default for complete page screenshots
  • 🎥 Animated WebP export - Save screencasts as high-quality animated WebP files
  • 💉 JavaScript injection - Execute custom JS before screencast capture
  • 📦 Minimal dependencies for fast npm installs
  • 🔌 MCP integration for seamless AI workflows
  • 🔋 Resource efficient - Automatic browser cleanup after 60 seconds of inactivity
  • 🧹 Memory management - Pages are closed after each screenshot to prevent leaks

Installation

Claude Code

claude mcp add screenshot-website-fast -s user -- npx -y @just-every/mcp-screenshot-website-fast

VS Code

code --add-mcp '{"name":"screenshot-website-fast","command":"npx","args":["-y","@just-every/mcp-screenshot-website-fast"]}'

Cursor

cursor://anysphere.cursor-deeplink/mcp/install?name=screenshot-website-fast&config=eyJzY3JlZW5zaG90LXdlYnNpdGUtZmFzdCI6eyJjb21tYW5kIjoibnB4IiwiYXJncyI6WyIteSIsIkBqdXN0LWV2ZXJ5L21jcC1zY3JlZW5zaG90LXdlYnNpdGUtZmFzdCJdfX0=

JetBrains IDEs

Settings → Tools → AI Assistant → Model Context Protocol (MCP) → Add

Choose "As JSON" and paste:

{"command":"npx","args":["-y","@just-every/mcp-screenshot-website-fast"]}

Raw JSON (works in any MCP client)

{
  "mcpServers": {
    "screenshot-website-fast": {
      "command": "npx",
      "args": ["-y", "@just-every/mcp-screenshot-website-fast"]
    }
  }
}

Drop this into your client's mcp.json (e.g. .vscode/mcp.json, ~/.cursor/mcp.json, or .mcp.json for Claude).

Prerequisites

  • Node.js 20.x or higher
  • npm or npx
  • Chrome/Chromium (automatically downloaded by Puppeteer)

Quick Start

MCP Server Usage

Once installed in your IDE, the following tools are available:

Available Tools

  • take_screenshot - Captures a high-quality screenshot of a webpage
    • Parameters:
      • url (required): The HTTP/HTTPS URL to capture
      • width (optional): Viewport width in pixels (max 1072, default: 1072)
      • height (optional): Viewport height in pixels (max 1072, default: 1072)
      • fullPage (optional): Capture full page screenshot with tiling (default: true)
      • waitUntil (optional): Wait until event: load, domcontentloaded, networkidle0, networkidle2 (default: domcontentloaded)
      • waitFor (optional): Additional wait time in milliseconds
      • directory (optional): Directory to save screenshots - returns file paths instead of base64 images

Usage Examples

Default usage (returns base64 images):

take_screenshot(url="https://example.com")

Save to directory (returns file paths):

take_screenshot(url="https://example.com", directory="/path/to/screenshots")

When using the directory parameter:

  • Screenshots are saved as PNG files with timestamps
  • File paths are returned instead of base64 data
  • For tiled screenshots, each tile is saved as a separate file
  • Directory is created automatically if it doesn't exist

take_screencast

Captures a series of screenshots over time to create a screencast. Only captures the top tile (1072x1072) of the viewport.

Parameters

  • url (required): The URL to capture
  • duration (optional): Total duration in seconds (default: 10)
  • interval (optional): Interval between screenshots in seconds (default: 2)
  • jsEvaluate (optional): JavaScript code to execute at the start
  • waitUntil (optional): Wait strategy: 'load', 'domcontentloaded', 'networkidle0', 'networkidle2'
  • waitForMS (optional): Additional wait time before starting
  • directory (optional): Save as animated WebP to directory (captures every 1 second)

Usage Examples

Basic screencast (5 frames over 10 seconds):

take_screencast(url="https://example.com")

Custom timing:

take_screencast(url="https://example.com", duration=15, interval=3)

With JavaScript execution:

take_screencast(
  url="https://example.com",
  jsEvaluate="document.body.style.backgroundColor = 'red';"
)

Save as animated WebP:

take_screencast(url="https://example.com", directory="/path/to/output")

When using the directory parameter:

  • An animated WebP is created with 1-second intervals
  • Individual frames are also saved as PNG files
  • The animation loops forever by default
  • WebP provides excellent quality:
    • Full color support (no 256 color limitation)
    • Efficient compression for web animations
    • Perfect for gradient backgrounds and smooth animations
    • Smaller file sizes compared to GIF with better quality

Development Usage

Install

npm install
npm run build

Capture screenshot

# Full page with automatic tiling (default)
npm run dev capture https://example.com -o screenshot.png

# Viewport-only screenshot  
npm run dev capture https://example.com --no-full-page -o screenshot.png

# Wait for specific conditions
npm run dev capture https://example.com --wait-until networkidle0 --wait-for 2000 -o screenshot.png

CLI Options

  • -w, --width <pixels> - Viewport width (max 1072, default: 1072)
  • -h, --height <pixels> - Viewport height (max 1072, default: 1072)
  • --no-full-page - Disable full page capture and tiling
  • --wait-until <event> - Wait until event: load, domcontentloaded, networkidle0, networkidle2
  • --wait-for <ms> - Additional wait time in milliseconds
  • -o, --output <path> - Output file path (required for tiled output)

Auto-Restart Feature

The MCP server includes automatic restart capability by default for improved reliability:

  • Automatically restarts the server if it crashes
  • Handles unhandled exceptions and promise rejections
  • Implements exponential backoff (max 10 attempts in 1 minute)
  • Logs all restart attempts for monitoring
  • Gracefully handles shutdown signals (SIGINT, SIGTERM)

For development/debugging without auto-restart:

# Run directly without restart wrapper
npm run serve:dev

Architecture

mcp-screenshot-website-fast/
├── src/
│   ├── internal/       # Core screenshot capture logic
│   ├── utils/          # Logger and utilities
│   ├── index.ts        # CLI entry point
│   ├── serve.ts        # MCP server entry point
│   └── serve-restart.ts # Auto-restart wrapper

Development

# Run in development mode
npm run dev capture https://example.com -o screenshot.png

# Build for production
npm run build

# Run tests
npm test

# Type checking
npm run typecheck

# Linting
npm run lint

Why This Tool?

Built specifically for AI vision workflows:

  1. Optimized for Claude Vision API - Automatic resolution limiting to 1072x1072 pixels (1.15 megapixels)
  2. Automatic tiling - Full pages split into perfect chunks for AI processing
  3. Always fresh - No caching ensures you get the latest content
  4. MCP native - First-class integration with AI development tools
  5. Simple API - Clean, straightforward interface for capturing screenshots

Contributing

Contributions are welcome! Please:

  1. Fork the repository
  2. Create a feature branch
  3. Add tests for new functionality
  4. Submit a pull request

Troubleshooting

Puppeteer Issues

  • Ensure Chrome/Chromium can be downloaded
  • Check firewall settings
  • Try setting PUPPETEER_SKIP_CHROMIUM_DOWNLOAD=true and provide custom executable

Screenshot Quality

  • Adjust viewport dimensions
  • Use appropriate wait strategies
  • Check if site requires authentication

Timeout Errors

  • Increase wait time with --wait-for flag
  • Use different --wait-until strategies
  • Check if site is accessible

License

MIT

README fetched from the repository 5 days ago. The original is authoritative.

Questions

About Website

How do I install Website?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Website safe to use with an AI agent?

Its trust score is 56 out of 100 (fair). It passes 5 of the 6 static security checks run so far; the failures are listed above. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Website still maintained?

The last commit was 11 months ago. That's long enough that you should check whether the maintainer is responding to issues before depending on it.