Provides visual I/O capabilities for macOS, allowing AI agents to capture, annotate, search, and manage screenshots natively on the operating system.
Enables AI agents to render visual diagrams from Mermaid code and open them for user review and approval flows.
Integrates with local Ollama instances to automate the naming, categorization, and semantic search indexing of captured visual content.
Snip
Visual communication layer between humans and AI agents for macOS.
Capture and annotate screenshots, render diagrams from code, review agent-generated visuals with approve/request-changes flow — all from the menu bar. AI organizes and indexes everything for semantic search. CLI and MCP integration let any AI agent use Snip as their visual I/O.
Install
brew install --cask rixinhahaha/snip/snipOr download the DMG directly from Releases (Apple Silicon only).
Quick Start (Development)
npm install
npm run rebuild # compile native modules
npm start # launch (tray icon appears in menu bar)Requires macOS 14+, Node.js 18+, and Xcode CLT (xcode-select --install). macOS 26+ recommended for native Liquid Glass effects.
For AI-powered organization, install Ollama separately. Snip detects your system Ollama and guides you through setup in Settings.
How It Works
Cmd+Shift+2 — Fullscreen overlay appears on whichever display the cursor is on, drag to select a region
Annotate — Rectangle, arrow, text, tag, blur brush, or AI segment tools
Esc — Copies annotated screenshot to clipboard
Cmd+S — Saves to disk + AI organizes in background
Screenshots saved to ~/Documents/snip/screenshots/. AI renames, categorizes, and indexes them for search.
Agent Integration (CLI & MCP)
Snip exposes a CLI and MCP server so AI agents can use it as their visual I/O:
# Render a Mermaid diagram and open for review
echo 'graph LR; A-->B-->C' | snip render --format mermaid --message "Does this flow look right?"
# Open an image for agent review
snip open screenshot.png --message "Is the layout correct?"The agent gets structured feedback: { status: "approved" | "changes_requested", edited, path, text? }. The user can annotate spatially, type text feedback, or just approve.
MCP tools: render_diagram, open_in_snip, search_screenshots, list_screenshots, get_screenshot, transcribe_screenshot, organize_screenshot, get_categories, install_extension.
Key Shortcuts
Shortcut | Action |
Cmd+Shift+2 | Capture screenshot |
Cmd+Shift+1 | Quick Snip (select & copy to clipboard) |
Cmd+Shift+S | Open semantic search |
Cmd+S | Save to disk (in editor) |
Esc / Enter | Copy to clipboard & close (in editor) |
V / R / T / A / G / B / S | Select / Rectangle / Text / Arrow / Tag / Blur / Segment tools |
U | Upscale image |
W | Transcribe text |
Documentation
Doc | Role | Contents |
Product Manager | Vision, feature specs, terminology, product principles | |
Designer | Color palettes (Dark/Light/Glass), component patterns, glass effects, icon specs | |
Developer | Code structure, conventions, IPC channels, data flow, key decisions | |
DevOps | Build pipeline, signing, native modules, environment setup | |
QA / PM | Detailed user flows for every feature, edge cases, test cases | |
Claude Code | Autonomous agent instructions, role references, documentation rules |
Tech Stack
Electron 33 / Fabric.js 7 / Mermaid.js 11 / Ollama (local LLM) / HuggingFace Transformers.js / SlimSAM (ONNX) / Chokidar 4 / electron-liquid-glass
On-Device Models
All AI runs locally — no cloud APIs needed for core features.
Model | Purpose | By | Link |
Vision LLM (naming, tagging, categorizing) | OpenBMB | ||
Object segmentation | Meta AI / Xenova | ||
Image upscaling (2x) | Conde et al. / Xenova | ||
Semantic search embeddings | Microsoft / Xenova | ||
Vision OCR | Text transcription | Apple | Built into macOS |
License
MIT