Screeny
A macOS-only server that enables LLMs to capture screenshots of specific application windows, providing visual context for development and debugging.

Screeny MCP Server: Privacy first macOS Screenshots for AI Agents
A privacy-first, macOS-only MCP server that enables AI agents to capture screenshots of pre-approved application windows, providing secure visual context for development and debugging tasks.
[!IMPORTANT] Requires Screen Capture permission + Window Approval Setup - see instructions below.
🔒 Privacy-First Design
Unlike other screenshot tools, Screeny requires explicit user approval for each window before it can be captured:
- Window approval system - Only pre-approved windows can be captured (approved during setup)
- User-controlled access - You decide exactly which windows are accessible
- Non-intrusive capture - Screenshots taken in background without changing window focus or interrupting your workflow
- No external connections - Screeny runs entirely on your device, screenshots are deleted immediately after use
Available Tools
-
listWindows- Lists all approved application windows available for screenshot capture.- Only shows user approved windows
-
takeScreenshot- Captures a screenshot of a specific window by its ID.- Captures windows in background - no need to bring window to front, but cannot capture minimized windows
- Provides actual pixel data - full-fidelity image, not OCR or text extraction
- JPEG compression with configurable cap - screenshots are always JPEG-compressed with a base64 payload cap (default preset: Medium / 250KB), configurable and clamped to 100–900KB
Resources
screeny://info- Server information and configuration details
Configuration
Claude Desktop
- Open Claude settings → Developer → Edit Config
- Add configuration
- Restart Claude Desktop after saving config
Using pipx
First install with: pipx install mcp-server-screeny
{
"mcpServers": {
"screeny": {
"command": "mcp-server-screeny",
"args": []
}
}
}
Note: If you get an
ENOENTerror, replace"mcp-server-screeny"with the full path to the executable (find it withwhich mcp-server-screenyin your terminal).
Using uvx
{
"mcpServers": {
"screeny": {
"command": "uvx",
"args": ["mcp-server-screeny"]
}
}
}
Note: If you get a "spawn uvx ENOENT" error, replace
"uvx"with the full path to uvx:which uvx # Find your uvx pathThen use that full path in the config (e.g.,
"/opt/homebrew/bin/uvx").
Cursor
- Open Cursor settings → Tools & Integrations → MCP Tools
- Add configuration
- Restart Cursor after saving config
Using pipx
First install with: pipx install mcp-server-screeny
{
"mcpServers": {
"screeny": {
"command": "mcp-server-screeny",
"args": []
}
}
}
Note: If you get an
ENOENTerror, replace"mcp-server-screeny"with the full path to the executable (find it withwhich mcp-server-screenyin your terminal).
Using uvx
{
"mcpServers": {
"screeny": {
"command": "uvx",
"args": ["mcp-server-screeny"]
}
}
}
Note: If you get a "spawn uvx ENOENT" error, replace
"uvx"with the full path to uvx:which uvx # Find your uvx pathThen use that full path in the config (e.g.,
"/opt/homebrew/bin/uvx").
[!NOTE] > Cursor MCP host
- Cursor auto context summarization can trigger looping tool calls (repeated screenshots).
- If this happens, lower the KB cap (use a smaller preset) or choose a model with a larger context window.
Setup
1. Grant Screen Capture Permission (Required)
Important: Grant permission before running window approval.
Note: You need to grant Screen Capture permission to BOTH:
- Your Terminal application (Terminal.app, iTerm2, etc.) - Required for running setup (can be disabled after)
- Your MCP host (Claude Desktop, Cursor) - Required for taking screenshots
To add them:
- Open System Settings > Privacy & Security > Screen & System Audio Recording
- Click the "+" button
- Add your Terminal application AND your MCP host application
- Restart both applications after granting permissions
2. Window Approval (Required)
After configuring your MCP client above, approve which windows can be captured.
If using pipx
# Interactive approval
mcp-server-screeny --setup
# Auto-approve all current windows
mcp-server-screeny --setup --allow-all
If using uvx
# Interactive approval
uvx mcp-server-screeny --setup
# Auto-approve all current windows
uvx mcp-server-screeny --setup --allow-all
Approvals are saved to ~/.screeny/approved_windows.json. Re-run setup when you want to update the list of approved windows.
Advanced Options (Optional)
During setup, you can configure the screenshot size preset (affects stability and clarity):
- Tiny (50KB) — most stable; fine text will blur
- Small (100KB) — recommended default; balanced clarity and stability
- Medium (250KB) — more detail; may be slower and heavier
- Large (500KB) — high detail; may trigger client summarization
- XL (750KB) — maximum detail; most error-prone
Your choice is saved in ~/.screeny/config.json as max_b64_kb. You can also override via the SCREENY_MAX_B64_KB environment variable. The active cap is clamped to 100–900KB.
Security & Privacy
- Only user-approved windows can be captured
- All processing stays local on your machine
- Screenshots are temporary and deleted immediately after use
Troubleshooting
Permission Issues
# Test window detection and permissions
mcp-server-screeny --debug
# Re-run setup if windows changed
mcp-server-screeny --setup
Common Issues
"spawn uvx ENOENT" error
- Solution: Use the full path to uvx in your MCP config instead of just
"uvx" - Find path with:
which uvx - Example:
"/opt/homebrew/bin/uvx"or"/usr/local/bin/uvx"
"No approved windows found"
- Solution: Run
mcp-server-screeny --setupfirst (oruvx mcp-server-screeny --setupif using uvx)
"Screen Recording permission required" or "No windows found"
- Solution: Grant Screen Recording permission in System Settings > Privacy & Security > Screen & System Audio Recording
- Click "+" button and manually add your MCP host (Claude Desktop, Cursor, etc.)
- Restart your MCP host application after granting permissions
- Try running setup again after granting permissions
Contributing
Pull requests are welcome! Feel free to contribute new ideas, bug fixes, or enhancements.
This is my first MCP project - if you encounter any bugs, please open an issue and I'll do my best to fix them!
Why I Built This
I created this tool to streamline my mobile development workflow. I was tired of manually taking screenshots repeatedly to describe UI issues. With Screeny, Cursor can directly capture screenshots of my iOS simulator and iterate on the design in a loop. I'm excited to see how others will use this!
Requirements
- Python 3.10+
- macOS
- Screen Capture permission
License
MIT License
相关服务器
Scout Monitoring MCP
赞助Put performance and error data directly in the hands of your AI assistant.
Alpha Vantage MCP Server
赞助Access financial market data: realtime & historical stock, ETF, options, forex, crypto, commodities, fundamentals, technical indicators, & more
LeetCode
Access LeetCode problems, user information, and contest data.
EOL MCP Server
Check software end-of-life (EOL) dates and support status using the endoflife.date API to provide accurate lifecycle and security information.
missiond
Multi-agent orchestration for Claude Code - spawn and control multiple Claude instances via MCP
Modellix Docs
Search the Modellix knowledge base to quickly find relevant technical information, code examples, and API references. Retrieve implementation details and official guides to solve development queries efficiently. Access direct links to documentation for deeper context on specific features and tools.
GoThreatScope
Go-based SBOM, vulnerability, and secret scanner with MCP support.
React Native AI Debugger
Enables AI assistants like Claude Code to capture logs, execute code, inspect state, and control navigation in your React Native app.
ReAPI OpenAPI
Serves multiple OpenAPI specifications to enable LLM-powered IDE integrations.
Adamik MCP Server
Interact with over 60 blockchain networks using any MCP client. Requires an Adamik API key.
Remote MCP Server (Authless)
An example of a remote MCP server deployable on Cloudflare Workers without authentication.
MCP Neurolora
An intelligent server for code analysis, collection, and documentation generation using the OpenAI API.