MCP Screenshot

Official

by Digital-Defiance · TypeScript

Screenshot capture with PII masking and cross-platform support for AI agents

Captures screenshots with PII masking and cross-platform support.

README.md19 KBView on GitHub

📸 MCP ACS Screenshot Server

Give AI agents visual superpowers to see, analyze, and document your applications like senior UX designers.

This enterprise-grade MCP server transforms AI from code-only assistants into visual experts capable of UI analysis, accessibility auditing, documentation generation, and responsive design testing.

🔗 Repository

This package is now maintained in its own repository: https://github.com/Digital-Defiance/mcp-screenshot

This repository is part of the AI Capabilitites Suite on GitHub.

🤔 Why Do AI Agents Need Visual Capabilities?

AI agents today are powerful but visually blind:

❌ Can read HTML/CSS but can't see actual layouts
❌ Can suggest UI improvements without seeing the real user experience
❌ Can't detect accessibility issues like poor contrast or spacing
❌ Can't create visual documentation or bug reports
❌ Can't analyze responsive design across different screen sizes

Result: You're stuck manually creating screenshots, documentation, and visual analysis that AI should handle.

🎯 Revolutionary Use Cases

📚 "AI, create professional documentation"

code

You: "Document the checkout process with screenshots"
AI: *Captures each step, analyzes UI flow*
AI: "Generated complete user guide with 6 annotated screenshots, highlighting key interactions and error states..."

🔍 "AI, audit this page for accessibility"

code

You: "Check if this form meets WCAG guidelines"
AI: *Captures form, analyzes visual elements*
AI: "Found 4 accessibility issues: contrast ratio 2.1:1 (needs 4.5:1), missing focus indicators, touch targets too small..."

🐛 "AI, create a detailed bug report"

code

You: "The mobile navigation is broken"
AI: *Captures different viewport sizes*
AI: "Created bug report with before/after screenshots, exact breakpoint where it fails, and CSS fix suggestions..."

🎨 "AI, compare these design variations"

code

You: "Which hero section design works better?"
AI: *Captures both versions, analyzes visual hierarchy*
AI: "Version B has 28% better visual flow—CTA more prominent, text hierarchy clearer, better use of whitespace..."

📱 "AI, test responsive design"

code

You: "How does this look on different screen sizes?"
AI: *Captures multiple viewport sizes*
AI: "Layout breaks at 768px—sidebar overlaps content. Here's the media query fix with visual proof..."

✨ What This Changes

Before: AI worked blind, relying on code descriptions

❌ "The button looks wrong" → AI guesses the issue
❌ "Create documentation" → AI writes generic text
❌ "Check accessibility" → AI only reviews code
❌ "Test responsive design" → AI can't see actual breakpoints

After: AI sees and analyzes your actual user interface

✅ Visual debugging - AI identifies exact pixel-level issues
✅ Smart documentation - AI creates guides with real screenshots and annotations
✅ Accessibility audits - AI measures actual contrast ratios and spacing
✅ Responsive testing - AI captures and compares different screen sizes
✅ Design analysis - AI evaluates visual hierarchy and user experience
✅ Professional reports - AI creates detailed visual evidence for bugs and improvements

🚀 Features

Multi-format Support: PNG, JPEG, WebP, BMP with configurable quality
Flexible Capture: Full screen, specific windows, or custom regions
Privacy Protection: PII masking with OCR-based detection for emails, phone numbers, and credit cards
Security Controls: Path validation, rate limiting, audit logging, and configurable policies
Cross-platform: Linux (X11/Wayland), macOS, Windows with native APIs
Multi-monitor Support: Capture from specific displays in multi-monitor setups
Enterprise Security: Window exclusion, audit logging, rate limiting
AI-Optimized: Structured responses perfect for AI agent workflows

Installation

NPM Installation

bash

npm install @ai-capabilities-suite/mcp-screenshot

System Requirements

Linux:

X11: imagemagick package (provides import command)
Wayland: grim package

bash

# Ubuntu/Debian
sudo apt-get install imagemagick grim

# Fedora
sudo dnf install ImageMagick grim

# Arch
sudo pacman -S imagemagick grim

macOS:

Built-in screencapture command (no additional dependencies)
Screen Recording permission required (System Preferences > Security & Privacy > Privacy > Screen Recording)

Windows:

No additional dependencies required

MCP Configuration

Add to your MCP settings file (e.g., ~/.kiro/settings/mcp.json or .kiro/settings/mcp.json):

json

{
  "mcpServers": {
    "screenshot": {
      "command": "node",
      "args": ["/path/to/mcp-screenshot/dist/cli.js"],
      "env": {
        "SCREENSHOT_ALLOWED_DIRS": "/home/user/screenshots,/tmp",
        "SCREENSHOT_MAX_CAPTURES_PER_MIN": "60",
        "SCREENSHOT_ENABLE_AUDIT_LOG": "true"
      }
    }
  }
}

🛠️ 5 Professional MCP Tools

Purpose-built for AI agents to capture, analyze, and work with visual information:

The server exposes 5 comprehensive MCP tools that enable AI agents to see and understand your applications:

1. screenshot_capture_full

Capture full screen or specific display.

Parameters:

display (string, optional): Display ID to capture (defaults to primary display)
format (string, optional): Image format - png, jpeg, webp, or bmp (default: png)
quality (number, optional): Compression quality 1-100 for lossy formats (default: 90)
savePath (string, optional): File path to save screenshot (returns base64 if not provided)
enablePIIMasking (boolean, optional): Enable PII detection and masking (default: false)

Example:

json

{
  "name": "screenshot_capture_full",
  "arguments": {
    "format": "png",
    "savePath": "/home/user/screenshots/desktop.png",
    "enablePIIMasking": true
  }
}

Response:

json

{
  "status": "success",
  "filePath": "/home/user/screenshots/desktop.png",
  "metadata": {
    "width": 1920,
    "height": 1080,
    "format": "png",
    "fileSize": 245678,
    "timestamp": "2024-12-01T10:30:00.000Z",
    "display": {
      "id": "0",
      "name": "Primary Display",
      "resolution": { "width": 1920, "height": 1080 },
      "position": { "x": 0, "y": 0 },
      "isPrimary": true
    },
    "piiMasking": {
      "emailsRedacted": 2,
      "phonesRedacted": 1,
      "creditCardsRedacted": 0,
      "customPatternsRedacted": 0
    }
  }
}

2. screenshot_capture_window

Capture specific application window by ID or title pattern.

Parameters:

windowId (string, optional): Window identifier (use windowId or windowTitle)
windowTitle (string, optional): Window title pattern to match (use windowId or windowTitle)
includeFrame (boolean, optional): Include window frame and title bar (default: false)
format (string, optional): Image format (default: png)
quality (number, optional): Compression quality 1-100 (default: 90)
savePath (string, optional): File path to save screenshot

Example:

json

{
  "name": "screenshot_capture_window",
  "arguments": {
    "windowTitle": "Chrome",
    "includeFrame": false,
    "format": "jpeg",
    "quality": 85
  }
}

Response:

json

{
  "status": "success",
  "data": "iVBORw0KGgoAAAANSUhEUgAA...",
  "mimeType": "image/jpeg",
  "metadata": {
    "width": 1280,
    "height": 720,
    "format": "jpeg",
    "fileSize": 89234,
    "timestamp": "2024-12-01T10:31:00.000Z",
    "window": {
      "id": "12345",
      "title": "Google Chrome",
      "processName": "chrome",
      "pid": 5678,
      "bounds": { "x": 100, "y": 100, "width": 1280, "height": 720 }
    }
  }
}

3. screenshot_capture_region

Capture specific rectangular region of the screen.

Parameters:

x (number, required): X coordinate of top-left corner
y (number, required): Y coordinate of top-left corner
width (number, required): Width of region in pixels
height (number, required): Height of region in pixels
format (string, optional): Image format (default: png)
quality (number, optional): Compression quality 1-100 (default: 90)
savePath (string, optional): File path to save screenshot

Example:

json

{
  "name": "screenshot_capture_region",
  "arguments": {
    "x": 100,
    "y": 100,
    "width": 800,
    "height": 600,
    "format": "png"
  }
}

Response:

json

{
  "status": "success",
  "data": "iVBORw0KGgoAAAANSUhEUgAA...",
  "mimeType": "image/png",
  "metadata": {
    "width": 800,
    "height": 600,
    "format": "png",
    "fileSize": 123456,
    "timestamp": "2024-12-01T10:32:00.000Z",
    "region": {
      "x": 100,
      "y": 100,
      "width": 800,
      "height": 600
    }
  }
}

4. screenshot_list_displays

List all connected displays with resolution and position information.

Parameters: None

Example:

json

{
  "name": "screenshot_list_displays",
  "arguments": {}
}

Response:

json

{
  "status": "success",
  "displays": [
    {
      "id": "0",
      "name": "Primary Display",
      "resolution": { "width": 1920, "height": 1080 },
      "position": { "x": 0, "y": 0 },
      "isPrimary": true
    },
    {
      "id": "1",
      "name": "Secondary Display",
      "resolution": { "width": 1920, "height": 1080 },
      "position": { "x": 1920, "y": 0 },
      "isPrimary": false
    }
  ]
}

5. screenshot_list_windows

List all visible windows with title, process, and position information.

Parameters: None

Example:

json

{
  "name": "screenshot_list_windows",
  "arguments": {}
}

Response:

json

{
  "status": "success",
  "windows": [
    {
      "id": "12345",
      "title": "Google Chrome",
      "processName": "chrome",
      "pid": 5678,
      "bounds": { "x": 100, "y": 100, "width": 1280, "height": 720 },
      "isMinimized": false
    },
    {
      "id": "67890",
      "title": "Terminal",
      "processName": "gnome-terminal",
      "pid": 9012,
      "bounds": { "x": 200, "y": 200, "width": 800, "height": 600 },
      "isMinimized": false
    }
  ]
}

Security Configuration

The server enforces security policies to control screenshot operations. Configure via environment variables or security policy file.

Environment Variables

SCREENSHOT_ALLOWED_DIRS: Comma-separated list of allowed directories for saving screenshots
SCREENSHOT_MAX_CAPTURES_PER_MIN: Maximum captures per minute (default: 60)
SCREENSHOT_ENABLE_AUDIT_LOG: Enable audit logging (default: true)
SCREENSHOT_BLOCKED_WINDOWS: Comma-separated list of window title patterns to exclude

Security Policy File

Create a security-policy.json file:

json

{
  "allowedDirectories": ["/home/user/screenshots", "/tmp/screenshots"],
  "blockedWindowPatterns": [
    ".*Password.*",
    ".*1Password.*",
    ".*LastPass.*",
    ".*Bitwarden.*",
    ".*Authentication.*"
  ],
  "maxCapturesPerMinute": 60,
  "enableAuditLog": true
}

Load the policy when starting the server:

typescript

import { MCPScreenshotServer } from "@ai-capabilities-suite/mcp-screenshot";
import * as fs from "fs";

const policy = JSON.parse(fs.readFileSync("security-policy.json", "utf-8"));
const server = new MCPScreenshotServer(policy);
await server.start();

Error Handling

All tools return structured error responses with error codes and remediation suggestions.

Error Codes

Code	Description	Remediation
`PERMISSION_DENIED`	Insufficient permissions to capture	Grant Screen Recording permission (macOS) or check user permissions
`INVALID_PATH`	File path outside allowed directories	Use a path within configured allowed directories
`WINDOW_NOT_FOUND`	Specified window does not exist	Use `screenshot_list_windows` to find available windows
`DISPLAY_NOT_FOUND`	Specified display does not exist	Use `screenshot_list_displays` to find available displays
`UNSUPPORTED_FORMAT`	Requested format not supported	Use png, jpeg, webp, or bmp
`CAPTURE_FAILED`	Screenshot capture failed	Check permissions and try again
`RATE_LIMIT_EXCEEDED`	Too many captures in time window	Wait before making additional requests
`INVALID_REGION`	Invalid region coordinates or dimensions	Ensure coordinates are non-negative and dimensions are positive
`OUT_OF_MEMORY`	Insufficient memory for operation	Reduce capture size or close other applications
`ENCODING_FAILED`	Image encoding failed	Try different format or reduce quality
`FILE_SYSTEM_ERROR`	File system operation failed	Check permissions and disk space

Error Response Format

json

{
  "status": "error",
  "error": {
    "code": "WINDOW_NOT_FOUND",
    "message": "Window with ID '12345' not found",
    "details": {
      "windowId": "12345"
    },
    "remediation": "Verify the window exists and is visible. Use screenshot_list_windows to see available windows."
  }
}

Troubleshooting

Linux Issues

Problem: import: command not found or grim: command not found

Solution: Install required packages:

bash

# X11
sudo apt-get install imagemagick

# Wayland
sudo apt-get install grim

Problem: Black screen or empty captures

Solution: Check display server environment variables:

bash

echo $DISPLAY  # Should show :0 or similar for X11
echo $WAYLAND_DISPLAY  # Should show wayland-0 or similar for Wayland

macOS Issues

Problem: PERMISSION_DENIED error

Solution: Grant Screen Recording permission:

Open System Preferences > Security & Privacy > Privacy
Select "Screen Recording" from the list
Add your terminal application or Node.js to the allowed list
Restart the application

Problem: Retina display captures are double resolution

Solution: This is expected behavior. Retina displays have 2x pixel density. Use the width and height from metadata to determine actual dimensions.

Windows Issues

Problem: Capture fails with access denied

Solution: Run the application with administrator privileges or check Windows Defender settings.

Problem: Multi-monitor captures show wrong display

Solution: Use screenshot_list_displays to get correct display IDs and positions.

General Issues

Problem: RATE_LIMIT_EXCEEDED error

Solution: The server limits captures to prevent abuse. Wait 60 seconds or adjust maxCapturesPerMinute in security policy.

Problem: INVALID_PATH error when saving

Solution: Ensure the save path is within allowed directories configured in security policy.

Problem: PII masking not working

Solution:

Ensure tesseract.js is properly installed
Check that eng.traineddata language file is available
PII masking requires OCR which may be slow on large images

Problem: Large file sizes

Solution:

Use JPEG format with lower quality (60-80) for smaller files
Use WebP format for best compression
Reduce capture region size if possible

Problem: Out of memory errors

Solution:

Capture smaller regions instead of full screen
Reduce quality settings
Close other applications to free memory
Use streaming for very large captures

Programmatic Usage

TypeScript/JavaScript

typescript

import { MCPScreenshotServer } from "@ai-capabilities-suite/mcp-screenshot";

// Create server with custom security policy
const server = new MCPScreenshotServer({
  allowedDirectories: ["/home/user/screenshots"],
  maxCapturesPerMinute: 30,
  enableAuditLog: true,
  blockedWindowPatterns: [".*Password.*"],
});

// Start server
await server.start();

// Server will handle MCP protocol requests via stdio
// Keep process running
process.on("SIGINT", async () => {
  await server.stop();
  process.exit(0);
});

Direct Capture Engine Usage

typescript

import { createCaptureEngine } from "@ai-capabilities-suite/mcp-screenshot";

// Create platform-specific capture engine
const engine = createCaptureEngine();

// Capture full screen
const fullScreen = await engine.captureScreen();

// List and capture windows
const windows = await engine.getWindows();
const window = windows.find((w) => w.title.includes("Chrome"));
if (window) {
  const buffer = await engine.captureWindow(window.id, false);
}

// Capture region
const region = await engine.captureRegion(100, 100, 800, 600);

// List displays
const displays = await engine.getDisplays();
console.log(`Found ${displays.length} displays`);

Development

This package is part of the AI Capabilities Suite monorepo.

Build

bash

npm run build

Test

bash

# Run all tests
npm test

# Run specific test suites
npm test -- capture
npm test -- security
npm test -- property

# Run with coverage
npm test -- --coverage

Project Structure

code

packages/mcp-screenshot/
├── src/
│   ├── capture/          # Platform-specific capture engines
│   ├── processing/       # Image processing and encoding
│   ├── privacy/          # PII detection and masking
│   ├── security/         # Security policy enforcement
│   ├── storage/          # File operations
│   ├── tools/            # MCP tool implementations
│   ├── interfaces/       # TypeScript interfaces
│   ├── types/            # Type definitions
│   ├── errors/           # Error classes
│   ├── server.ts         # MCP server implementation
│   └── cli.ts            # CLI entry point
├── README.md
├── TESTING.md
└── package.json

Contributing

Contributions are welcome! Please ensure:

All tests pass (npm test)
Code follows TypeScript best practices
New features include tests and documentation
Security considerations are addressed

License

MIT

Support

For issues and questions:

GitHub Issues: Create an issue
Documentation: See TESTING.md for testing guide
Security: Report security issues privately to security@example.com

Install

Add to VS Code Add to Cursor

claude_desktop_config.json

json

{
  "mcpServers": {
    "mcp-screenshot": {
      "command": "npx",
      "args": [
        "-y",
        "@ai-capabilities-suite/mcp-screenshot@1.0.3"
      ]
    }
  }
}

GitHub

Stars: 2
Forks: 0
Open issues: 0
Language: TypeScript
Default branch: main
Latest release: v1.5.23 · 12/12/2025
Created: 12/3/2025
Updated: 1/7/2026

Trust grades

License: -
Maintenance: C
Quality: Not tested yet

Graded A-F from license permissiveness and commit recency; a cross-model quality score is coming soon.

npm

Package: @ai-capabilities-suite/mcp-screenshot
Latest version: 1.5.23
Published: 12/12/2025

Details

Name: io.github.Digital-Defiance/mcp-screenshot
Version: 1.0.3
Status: active
Published: 12/5/2025
Updated: 12/5/2025
Official: Yes
Repository: Digital-Defiance/mcp-screenshot
Registry record: View server.json

Connect agentage Memory

The cross-vendor memory every AI reads and writes - Claude, Cursor, ChatGPT - mirrored as plain markdown files you own and can export anytime.

MCP endpointmemory.agentage.io/mcp

Connect Memory

One memory. Every AI. Owned by you.