MCP 服务器

Snip

Screenshot and diagram tool for AI agents. Capture and annotate screenshots to show Claude what you mean — or let the agent render Mermaid diagrams and open them for visual review. Approve, annotate, or request changes with text feedback. Built-in review mode with structured responses. CLI and MCP server for Claude Code, Cursor, Windsurf, Cline. macOS, open source, free.

README

Snip

snipit.dev

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/snip

Or 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
`docs/PRODUCT.md`	Product Manager	Vision, feature specs, terminology, product principles
`docs/DESIGN.md`	Designer	Color palettes (Dark/Light/Glass), component patterns, glass effects, icon specs
`docs/ARCHITECTURE.md`	Developer	Code structure, conventions, IPC channels, data flow, key decisions
`docs/DEVOPS.md`	DevOps	Build pipeline, signing, native modules, environment setup
`docs/USER_FLOWS.md`	QA / PM	Detailed user flows for every feature, edge cases, test cases
`CLAUDE.md`	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
MiniCPM-V	Vision LLM (naming, tagging, categorizing)	OpenBMB	HF
SlimSAM-77-uniform	Object segmentation	Meta AI / Xenova	HF
Swin2SR-lightweight-x2-64	Image upscaling (2x)	Conde et al. / Xenova	HF
all-MiniLM-L6-v2	Semantic search embeddings	Microsoft / Xenova	HF
Vision OCR	Text transcription	Apple	Built into macOS