snip

rixinhahahaMIT更新放缓

一款人机视觉沟通工具,可截图、标注、渲染 Mermaid 图表,并借助 Ollama 做语义搜索整理,适合把屏幕内容快速交给 AI 理解。

The visual communication layer between humans and AI agents. Capture, annotate, render diagrams, and organize with AI — powered by Electron and Ollama. macOS & Linux.

v1.3.14发布于 2026-05-07 21:45 UTC352star33fork同步于 2026-11-07 07:33 UTC

怎么上手

由 README 提炼,执行前请确认
brew install --cask rixinhahaha/snip/snip
snip setup
硬件要求按 README 说明

README 未说明显存要求;AI 功能依赖本地 Ollama 视觉模型,需自行安装 Ollama。系统要求 macOS 14+ 或 Linux(Wayland),开发需 Node.js 18+。

支持 Apple Silicon
下载:源码 zip·GitHub Releases(安装包)本站不镜像安装包

项目说明

由 README 自动翻译提炼

给 AI 编码助手加上视觉输出层,让 Claude Code 直接渲染图表和预览供你批注

一款人机视觉沟通工具,可截图、标注、渲染 Mermaid 图表,并借助 Ollama 做语义搜索整理,适合把屏幕内容快速交给 AI 理解。

核心能力

  • 用 snip render 把 Mermaid 或 HTML 渲染成可审阅窗口
  • 在截图上圈画、加箭头和文字,把批注回传给 AI
  • 返回 approved 或 changes_requested 的结构化 JSON
  • 内置 MCP server,供无 shell 权限的 agent 调用
  • 本地 Ollama 视觉模型自动命名、分类并支持语义搜索截图

容易踩的坑

AI 分类和语义搜索需要额外安装并运行 Ollama,否则相关功能不可用;macOS 版仅提供 Apple Silicon 安装包;Linux 需 Wayland 环境,X11 未提及支持。

内容由模型根据 README 生成,可能不完整或不准确,以官方文档为准。看原文 README

最近发版

5 个版本v1.3.14
  1. v1.3.141.3.14最新正式版2026-05-07 · 2026-05-07 21:45 UTC
  2. v1.3.131.3.132026-04-11 · 2026-04-11 04:57 UTC
    查看更新说明

    What's Changed

    Full Changelog: https://github.com/rixinhahaha/snip/compare/v1.3.12...v1.3.13

  3. v1.3.121.3.122026-04-10 · 2026-04-10 23:23 UTC
    查看更新说明

    What's Changed

    Full Changelog: https://github.com/rixinhahaha/snip/compare/v1.3.11...v1.3.12

  4. v1.3.111.3.112026-04-05 · 2026-04-05 03:55 UTC
    查看更新说明
  5. v1.3.101.3.102026-04-04 · 2026-04-04 13:43 UTC
    查看更新说明

用户点评

登录用户的真实使用记录,可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译

    Snip

    Visual mode for Claude Code. Also works with Cursor, Windsurf, Cline, and any AI coding CLI.

    Install Snip, run snip setup, and your AI coding agent starts rendering diagrams and previews instead of describing them in text. Architecture? Sequence diagram. Data model? ER diagram. UI component? Rendered preview. All visual, all automatic.

    You review what Claude renders — approve it, or annotate directly on screen. Circle what's wrong, add an arrow, write a note. Claude sees your annotations and iterates.

    How It Works

    Snip runs as a menu bar / system tray app with a CLI that any AI coding agent can call directly:

    # Claude writes a Mermaid diagram to a file, then renders it for your review
    snip render --format mermaid < architecture.mmd
    
    # Claude renders an HTML component preview
    snip render --format html < preview.html
    
    # Open any image for annotated review
    snip open screenshot.png --message "Is the layout correct?"
    
    # Search your screenshot library
    snip search "login page error"
    

    The agent gets structured JSON back: { status: "approved" | "changes_requested", edited, path, text }. You can annotate spatially, type text feedback, or just approve.

    Built for Claude Code. Also works with Cursor, Windsurf, Cline, and anything that can run a shell command. Ships an MCP server for agents without shell access.

    Install

    macOS (Homebrew)

    brew install --cask rixinhahaha/snip/snip
    

    Or download the DMG from Releases (Apple Silicon).

    Linux

    Download from Releases:

    • AppImage (portable, any distro) — Snip-x.y.z-x86_64.AppImage
    • deb (Ubuntu/Debian) — Snip-x.y.z-amd64.deb

    Use with Claude Code

    snip setup
    

    That's it. snip setup configures Claude Code to use visual output automatically — rules, the /diagram skill, and permissions. Next time you ask Claude about architecture, data models, or flows, it renders a diagram instead of describing it in text.

    Try /diagram in any Claude Code conversation to visualize what you're discussing.

    CLI Reference

    CommandWhat it does
    snip setupEnable visual mode for Claude Code (and Cursor, Windsurf, Cline)
    snip render --format mermaidRender Mermaid diagram from stdin, open for review
    snip render --format htmlRender HTML from stdin, open for review
    snip open <path>Open any image for annotated review
    snip search <query>Search screenshot library by description
    snip transcribe <path>Extract text from an image via OCR
    snip listList all saved screenshots with metadata
    snip get <path>Get metadata for a specific screenshot
    snip organize <path>Queue screenshot for AI categorization
    snip categoriesList all categories

    All review commands (render, open) block until the user finishes and return structured JSON.

    MCP Server

    For agents without shell access (Claude Desktop, hosted environments), Snip also ships an MCP server:

    {
      "mcpServers": {
        "snip": {
          "command": "node",
          "args": ["/path/to/snip/src/mcp/server.js"]
        }
      }
    }
    

    The MCP server exposes the same capabilities: render_diagram, open_in_snip, search_screenshots, list_screenshots, get_screenshot, transcribe_screenshot, organize_screenshot, get_categories, install_extension.

    Also a Screenshot Tool

    Snip is a full screenshot + annotation app on its own:

    • Cmd+Shift+2 (macOS) / Ctrl+Shift+2 (Linux) — Capture with region select or window snap
    • Cmd+Shift+1 / Ctrl+Shift+1 — Quick Snip (capture straight to clipboard)
    • Annotate — Rectangle, arrow, text, tag, blur brush, AI segment
    • Esc — Copy to clipboard and close
    • Cmd+S — Save to disk + AI organizes in background

    AI organization uses a local vision LLM (via Ollama) to name, categorize, and tag every saved screenshot. Semantic search finds any screenshot by describing what was in it.

    Key Shortcuts

    On Linux, replace Cmd with Ctrl.

    ShortcutAction
    Cmd+Shift+2Capture screenshot
    Cmd+Shift+1Quick Snip (select & copy to clipboard)
    Cmd+Shift+SOpen semantic search
    Cmd+SSave to disk (in editor)
    Esc / EnterCopy to clipboard & close (in editor)
    V / C / R / T / A / G / B / SSelect / Crop / Rectangle / Text / Arrow / Tag / Blur / Segment tools
    Cmd+ZUndo (annotations first, then crop, then segment cutout)

    Development

    npm install
    npm run rebuild   # compile native modules (macOS)
    npm start         # launch (tray icon appears)
    

    Requires macOS 14+ or Linux (Wayland), Node.js 18+.

    Documentation

    DocContents
    docs/PRODUCT.mdVision, feature specs, terminology
    docs/DESIGN.mdColor palettes, component patterns, glass effects
    docs/ARCHITECTURE.mdCode structure, IPC channels, data flow
    docs/DEVOPS.mdBuild pipeline, signing, native modules
    docs/USER_FLOWS.mdStep-by-step flows, edge cases, test cases

    Tech Stack

    Electron 33 / Fabric.js 7 / Mermaid.js 11 / Ollama (local LLM) / HuggingFace Transformers.js / SlimSAM (ONNX)

    All AI runs locally — no cloud APIs needed for core features.

    Sponsor

    Snip is free, open source, and built by two developers. If it's useful to your workflow, consider sponsoring the project.

    Sponsorships help cover:

    • Apple Developer Program — required for code signing and notarization so macOS doesn't block the app
    • CI/CD costs — automated builds, testing, and release pipelines for macOS and Linux
    • Continued development — new annotation tools, better AI integration, Linux improvements, and bug fixes

    Every contribution helps keep Snip maintained and improving. Thank you!

    License

    MIT