PeepIt MCP: Lightning-fast macOS Screenshots for AI Agents

PeepIt: Because Your AI Deserves to See What You See

Ever wish your AI assistant could just look at your screen and get it? PeepIt is here to grant your digital sidekick the gift of sight—no magic wands required. Whether you're debugging a UI, capturing a bug in the wild, or just want to know what's lurking behind that mysterious window, PeepIt's got your back (and your screen).

What is PeepIt?

PeepIt is a macOS-only MCP server that lets AI agents capture screenshots of your apps, windows, or the whole system—then analyze them with local or cloud-based AI models. It's like giving your AI a pair of glasses and a magnifying glass, all in one.

• Capture screenshots of anything: the whole screen, a single app, or that one window you can never find

• Analyze visual content with AI vision models (local or cloud—your call)

• List running apps and windows for laser-targeted captures

• Work non-intrusively—no window focus stealing, no workflow interruptions, no drama

Key Features

• 🚀 Fast & Non-intrusive: Blink and you'll miss it—PeepIt uses Apple's ScreenCaptureKit for lightning-fast screenshots, all without hijacking your window focus or interrupting your groove.

• 🎯 Smart Window Targeting: Fuzzy matching so sharp, it'll find the right window even if you only remember half its name (we've all been there).

• 🤖 AI-Powered Analysis: Ask questions about your screenshots and get answers from GPT-4o, Claude, or local models—because sometimes you need a second set of (robotic) eyes.

• 🔒 Privacy-First: Prefer to keep things on the down-low? Run everything locally with Ollama, or call in the cloud cavalry only when you really need it.

• 📦 Easy Installation: One-click install via Cursor, or just a quick npm/npx incantation—no arcane rituals required.

• 🛠️ Developer-Friendly: Clean JSON API, TypeScript support, and logs so comprehensive you'll wonder if PeepIt is secretly writing your memoirs.

Installation

Requirements

• macOS 14.0+ (Sonoma or later)

• Node.js 20.0+

• Screen Recording Permission (don't worry, you'll be prompted—no need to go spelunking in System Settings)

Quick Start

For Cursor IDE

Or manually add to your Cursor settings:

For Claude Desktop

Edit your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the PeepIt configuration (copy, paste, and you're halfway to AI vision):

Then restart Claude Desktop. (Yes, you really do have to restart it. We checked.)

Configuration

PeepIt is as configurable as your favorite text editor. Use environment variables to tune it to your workflow:

Available Environment Variables

AI Provider Configuration

The PEEPIT_AI_PROVIDERS variable is your golden ticket to AI-powered screenshot analysis. Want PeepIt to answer questions about your screen? Just list your favorite models:

PEEPIT_AI_PROVIDERS="openai/gpt-4o,ollama/llava:latest,anthropic/claude-3-haiku-20240307"

Or, if you're a semicolon connoisseur:

PEEPIT_AI_PROVIDERS="openai/gpt-4o;ollama/llava:latest;anthropic/claude-3-haiku-20240307"

Each entry is provider_name/model_identifier. Supported providers: ollama (for local), openai (for the cloud), and soon, anthropic (for the truly adventurous).

PeepIt will try providers in order, checking for API keys or local services as needed. You can override the model per request if you're feeling particular.

Setting Up Local AI with Ollama

Ollama brings AI vision to your desktop—no cloud required, no data leaving your Mac. (Your secrets are safe. Probably.)

Installing Ollama

Downloading Vision Models

For beefy machines:

For lighter laptops:

Model Size Guide:

• qwen2-vl:7b - ~4GB download, ~6GB RAM (great for mortals)

• llava:7b - ~4.5GB download, ~8GB RAM

• llava:13b - ~8GB download, ~16GB RAM

• llava:34b - ~20GB download, ~40GB RAM (bring snacks)

Configuring PeepIt with Ollama

Add Ollama to your Claude Desktop config:

For lighter machines:

Mix and match AI providers:

macOS Permissions

PeepIt needs a few macOS permissions to work its magic. Don't worry, it's not asking for your Netflix password.

1. Screen Recording Permission (Required)

macOS Sequoia (15.0+):

1. System Settings → Privacy & Security

2. Scroll to Screen & System Audio Recording

3. Toggle on your terminal or MCP client

4. Restart the app (yes, again)

macOS Sonoma (14.0) and earlier:

1. System Preferences → Security & Privacy → Privacy

2. Select Screen Recording

3. Click the lock, enter your password

4. Add your terminal or MCP client

5. Restart the app

Apps that need permission:

• Terminal.app

• Claude Desktop

• VS Code

• Cursor

2. Accessibility Permission (Optional, but nice)

macOS Sequoia (15.0+):

1. System Settings → Privacy & Security → Accessibility

2. Toggle on your terminal/MCP client

macOS Sonoma (14.0) and earlier:

1. System Preferences → Security & Privacy → Privacy

2. Select Accessibility

3. Add your terminal/MCP client

Testing & Debugging

Using MCP Inspector

Want to see PeepIt in action? Fire up the MCP Inspector:

Direct CLI Testing

Expected output:

Available Tools

PeepIt gives you three main tools—think of them as your AI's Swiss Army knife:

1. image - Capture Screenshots

Snap a screenshot of your Mac—screen, app, or window. Shadows and frames? Gone. (You're welcome.)

Note: Screen captures are always saved to files (no Base64 for giant images—your stack won't like it). If you ask for format: "data", PeepIt will politely ignore you and save a PNG instead, with a gentle warning.

Examples:

Browser Helper Filtering: PeepIt is smart enough to avoid browser helper processes (no more "Google Chrome Helper (Renderer)" shenanigans). You'll get the real browser window, or a clear message if it's not running.

File Naming and Path Behavior:

• Single capture? Your path is used as-is.

• Multiple captures? PeepIt adds metadata to filenames so nothing gets overwritten.

• Directory path? PeepIt generates unique names for you.

• Long filenames? PeepIt trims them to fit macOS's 255-byte limit, keeping your emoji and non-Latin scripts intact.

• Invalid formats? Only PNG and JPEG are allowed. Anything else gets converted, with a friendly warning.

2. list - System Information

List running apps, windows, or check server status. Because sometimes you just need to know what's out there.

Examples:

3. analyze - AI Vision Analysis

Feed an image to your AI and ask it anything. (Well, almost anything.)

Examples:

Testing

PeepIt comes with tests galore:

TypeScript Tests

• Unit Tests: For the code that likes to be alone

• Integration Tests: For the code that plays well with others

• Platform-Specific Tests: Some tests need macOS and the Swift binary

Swift Tests

Platform Support

• macOS: All tests

• Linux/CI: TypeScript-only (Swift tests are skipped)

• Env Vars:

  • SKIP_SWIFT_TESTS=true: Skip Swift tests

  • CI=true: Skip Swift tests automatically

Troubleshooting

Debug Mode

Getting Help

• 📚 Documentation

Building from Source

Development Setup

Local Development Configuration

For local dev:

Or, running directly with node:

Use absolute paths and unique server names to avoid confusion.

AppleScript Version (Legacy)

For the old-school crowd:

Note: No AI analysis or MCP features in this version.

Manual Configuration for Other MCP Clients

Tool Documentation

image - Screenshot Capture

Capture your Mac's screen and optionally analyze it. Shadows and frames are automatically banished.

Parameters:

• app_target (string, optional): Specifies the capture target. If omitted or empty, captures all screens.

  • Examples:

    • "screen:INDEX": Captures the screen at the specified zero-based index (e.g., "screen:0"). (Note: Index selection from multiple screens is planned for full support in the Swift CLI).

    • "frontmost": Captures the frontmost window of the currently active application.

    • "AppName": Captures all windows of the application named AppName (e.g., "Safari", "com.apple.Safari"). Fuzzy matching is used.

    • "PID:ProcessID": Captures all windows of the application with the specified process ID (e.g., "PID:663"). Useful when multiple instances of the same app are running.

    • "AppName:WINDOW_TITLE:Title": Captures the window of AppName that has the specified Title (e.g., "Notes:WINDOW_TITLE:My Important Note").

    • "AppName:WINDOW_INDEX:Index": Captures the window of AppName at the specified zero-based Index (e.g., "Preview:WINDOW_INDEX:0" for the frontmost window of Preview).

• path (string, optional): Base absolute path for saving the captured image(s). If format is "data" and path is also provided, the image is saved to this path (as a PNG) AND Base64 data is returned. If a question is provided and path is omitted, a temporary path is used for capture, and the file is deleted after analysis.

• question (string, optional): If provided, the captured image will be analyzed. The server automatically selects an AI provider from those configured in the PEEPIT_AI_PROVIDERS environment variable.

• format (string, optional, default: "png"): Specifies the output image format or data return type.

  • "png" or "jpg": Saves the image to the specified path in the chosen format. For application captures: if path is not provided, behaves like "data". For screen captures: always saves to file.

  • "data": Returns Base64 encoded PNG data of the image directly in the MCP response. If path is also specified, a PNG file is also saved to that path. Note: Screen captures cannot use this format and will automatically fall back to PNG file format.

  • Invalid values (empty strings, null, or unrecognized formats) automatically fall back to "png".

• capture_focus (string, optional, default: "background"): Controls window focus behavior during capture.

  • "background": Captures without altering the current window focus (default).

  • "foreground": Attempts to bring the target application/window to the foreground before capture. This might be necessary for certain applications or to ensure a specific window is captured if multiple are open.

Behavior with

• If a question is provided, the tool will capture the image (saving it to path if specified, or a temporary path otherwise).

• This image is then sent to an AI model for analysis. The AI provider and model are chosen automatically by the server based on your PEEPIT_AI_PROVIDERS environment variable (trying them in order until one succeeds).

• The analysis result is returned as analysis_text in the response. Image data (Base64) is NOT returned in the content array when a question is asked.

• If a temporary path was used for the image, it's deleted after the analysis attempt.

Output Structure (Simplified):

• content: Can contain ImageContentItem (if format: "data" or path was omitted, and no question) and/or TextContentItem (for summaries, analysis text, warnings).

• saved_files: Array of objects, each detailing a file saved to path (if path was provided).

• analysis_text: Text from AI (if question was asked).

• model_used: AI model identifier (if question was asked).

For detailed parameter documentation, see docs/spec.md.

File Naming and Path Behavior

PeepIt intelligently manages output paths to prevent file overwrites while respecting your intentions:

Key Principle: Single vs Multiple Captures

When you provide a specific file path (e.g., ~/Desktop/screenshot.png), PeepIt determines whether to use it exactly or add metadata based on the capture context:

1. Single Capture → Exact Path

   • Capturing one specific window

   • Capturing one specific screen (when only one display exists)

   • Capturing with app_target: "frontmost"

   • Your path is used exactly as specified

2. Multiple Captures → Metadata Added

   • Capturing all windows of an app (mode: "multi" or multiple windows exist)

   • Capturing all screens (when multiple displays exist)

   • Capturing with no specific target (defaults to all screens)

   • Metadata is appended to prevent overwrites

Examples:

Long Filename Protection:

PeepIt automatically handles filesystem limitations:

• Truncates filenames exceeding macOS's 255-byte limit

• Preserves UTF-8 multibyte characters (emoji, non-Latin scripts)

• Ensures metadata is always included when needed

• Never creates invalid filenames

Example:

Format Validation:

• Invalid formats ("bmp", "gif", "tiff", etc.) automatically convert to PNG

• You'll receive a clear warning message when format correction occurs

• Only "png" and "jpg"/"jpeg" are valid formats

Browser Helper Filtering:

PeepIt automatically filters out browser helper processes when searching for common browsers (Chrome, Safari, Firefox, Edge, Brave, Arc, Opera). This prevents confusing errors when helper processes like "Google Chrome Helper (Renderer)" are matched instead of the main browser application.

Examples:

Browser-Specific Error Messages:

• Instead of generic "Application not found"

• Shows clear messages like "Chrome browser is not running or not found"

• Only applies to browser identifiers - other apps work normally

Technical Features

• Multi-display support: Each monitor gets its own moment in the spotlight

• Smart app targeting: Fuzzy matching for app names

• Multiple formats: PNG, JPEG, WebP, HEIF

• Automatic naming: Timestamp-based, no overwrites

• Permission checking: No surprises

• Application listing: See what's running

• Window enumeration: List all windows for an app

• PID targeting: For the process-obsessed

• Status monitoring: Know what's active

• Provider agnostic: Ollama, OpenAI, and soon Anthropic

• Natural language: Ask questions about images

• Configurable: Environment-based

• Fallback support: Automatic failover between providers

Architecture

Technical Details

JSON Output Format

The Swift CLI outputs structured JSON when called with --json-output:

MCP Integration

The Node.js server provides:

• Schema validation via Zod

• Proper MCP error codes

• Structured logging via Pino

• Full TypeScript type safety

Security

PeepIt respects macOS security:

• Checks permissions before operations

• Graceful handling of missing permissions

• Clear guidance for permission setup

Development

Testing Commands

Building

Known Issues

• FileHandle warning: Non-critical Swift warning about TextOutputStream conformance

• AI Provider Config: Requires PEEPIT_AI_PROVIDERS environment variable for analysis features

License

MIT License - see LICENSE file for details.

Contributing

1. Fork the repo

2. Create a feature branch (git checkout -b feature/amazing-feature)

3. Commit your changes (git commit -m 'Add amazing feature')

4. Push to the branch (git push origin feature/amazing-feature)

5. Open a Pull Request

Author

Created by Peter Steinberger - @mantisware

Read more about PeepIt's design and implementation in the blog post.