ScreenCapture MCP Server

Give your AI assistant eyes. Capture and analyze your screen directly from Cursor, Claude Code, or any MCP-compatible editor.

---

An MCP (Model Context Protocol) server that lets AI assistants autonomously capture and analyze your screen — no manual screenshots, no copy-paste, no uploads. Just ask.

You:  "Capture my screen and review the UI"
AI:   *captures screen, analyzes colors, layout, spacing, and gives actionable feedback*

Why?

Every time you want AI help with a design, you have to: screenshot, save, upload, wait. This eliminates all of that. Your AI assistant can now see exactly what you see, instantly.

---

Features

10 tools your AI assistant can use autonomously:

Tool	What It Does
capture_screen	Full screen or specific monitor capture
capture_region	Capture any rectangular area (x, y, w, h)
capture_active_window	Capture just the focused window
analyze_design	Full design DNA extraction (colors, layout, typography, spacing)
extract_color	Get exact color at any screen coordinate (hex, RGB, HSV)
analyze_color_palette	Find dominant colors in any region
check_contrast	WCAG accessibility contrast ratio checker
start_continuous_capture	Time-lapse capture with optional OCR text extraction
get_monitor_info	List all monitors with resolutions
cleanup_logs	Maintenance — remove old log files

4 built-in prompts for common workflows:

• review_ui — Comprehensive UI/UX review
• check_accessibility — WCAG compliance audit
• quick_design_check — Quick design issues scan
• color_analysis — Color usage and contrast analysis

---

Quick Start

Prerequisites

• Python 3.10+
• An MCP-compatible editor (Cursor, Claude Code, etc.)

Install

git clone https://github.com/xDarkzx/ScreenCapture-MCP.git
cd ScreenCapture-MCP
pip install -r requirements.txt

Configure Your Editor

Cursor

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "screen-access": {
      "command": "python",
      "args": ["-m", "src"],
      "cwd": "/path/to/ScreenCapture-MCP"
    }
  }
}

Restart Cursor. Look for "screen-access" with a green dot in the MCP panel.

Claude Code

claude mcp add screen-access -- python -m src --cwd /path/to/ScreenCapture-MCP

Other MCP Editors

Point your MCP config to:

• Command: python
• Args: ["-m", "src"]
• Working directory: the cloned repo folder

Test It

Ask your AI: "Capture my screen and tell me what you see"

---

Usage Examples

Design Review

"Capture my screen and review the spacing and alignment"

Accessibility Audit

"Check if the text on my landing page meets WCAG contrast standards"

Color Extraction

"What are the exact colors used in this design?"

Code from Screenshot

"Capture this landing page and recreate it in HTML/CSS"

Continuous Capture (for scrolling content)

"Start continuous capture for 30 seconds while I scroll through this documentation"

Multi-Monitor

"Capture my second monitor and analyze the layout"

---

Architecture

┌─────────────────┐    MCP/stdio     ┌──────────────────────┐
│   AI Editor     │ ◄──────────────► │  ScreenCapture MCP   │
│  (Cursor, etc.) │                  │  Server (Python)     │
└─────────────────┘                  └──────────┬───────────┘
                                                │
                                     ┌──────────▼───────────┐
                                     │  Capture Engine      │
                                     │  - mss (screen grab) │
                                     │  - Pillow (analysis) │
                                     │  - OCR (optional)    │
                                     └──────────────────────┘
                                                │
                                         In-memory only
                                        (no disk writes)

Privacy by design:

• 100% local — nothing leaves your machine
• In-memory only — no screenshots saved to disk
• No telemetry — zero data sent anywhere
• Open source — audit the code yourself

---

Dependencies

Package	Purpose
mcp	Model Context Protocol SDK
mss	Fast cross-platform screen capture
Pillow	Image processing and analysis
pywin32	Windows window detection (Windows only)
easyocr	Optional — text extraction from screenshots
numpy	Required by easyocr

---

Troubleshooting

Problem	Solution
MCP showing red/yellow	Check the path in your config, make sure Python is installed
MCP not appearing	Fully restart your editor, verify JSON config syntax
"No module named src"	Make sure cwd points to the repo root
OCR not working	Install easyocr: pip install easyocr (optional feature)
Active window capture fails	Windows only — falls back to full screen on other platforms

Check mcp_server.log in the project folder for detailed error messages.

---

Contributing

Contributions welcome! Open an issue or submit a PR.

License

MIT

---

Built for the MCP ecosystem — works with any editor that supports the Model Context Protocol.