wsnap

macOS-style screen capture for Windows.

Press Shift+F1, drag a region, and the image is already on your clipboard. Pick an action from the toolbar that appears right at your selection — copy, save, edit, OCR, GIF, pin — or just drag the floating thumbnail straight into any app as a real file.

Native. Offline. No account, no tracking. GPL-3.0.

---

Why wsnap

Most Windows capture tools make you fish a file out of a folder, or they copy a file path to the clipboard when you actually wanted the image. wsnap treats the two things people do with a screenshot every day — paste it and drag it somewhere — as first-class:

• The capture is on your clipboard as an image the instant you release the mouse. Paste it anywhere with Ctrl+V.
• The floating thumbnail is a real, draggable file. Drop it into Slack, a chat, an editor, an email — and it stays put so you can drop it again somewhere else.

Everything else — a precise frozen-screen overlay, a pixel loupe with HEX color readout, an annotation editor, on-device OCR, GIF recording, scrolling capture — is built around that core in one consistent dark UI, running quietly from the tray.

Features

• Capture → floating thumbnail → drag-and-drop. Thumbnails stack at the bottom-right, up to a configurable number.
• Clipboard-first. Click = copy the image; auto-copy on capture (optional) so it's ready before you even click. Ctrl+Click = copy the file path.
• Action toolbar at your selection — Copy · Save · Edit · OCR · GIF · Pin (keys C / Enter / E / T / G / P, Esc to cancel).
• Precise overlay. Freezes the screen, brightens only your selection (punch-through dim), shows live W×H, and a magnifier loupe with pixel coordinates and the HEX color under the cursor. Physical-pixel cursor grab — correct on mixed-DPI multi-monitor setups.
• Color picker (eyedropper). Click any pixel → #RRGGBB copied.
• Annotation editor — arrow, line, rectangle, ellipse, pen, highlighter, text, numbered steps, mosaic / blur redaction, crop. Pick thickness and color, select / move / delete objects (V), undo and redo, undoable crop, copy to clipboard (Ctrl+C), Shift to constrain (45° / square). Keyboard-driven.
• Great on-device OCR (KO + EN). PaddleOCR PP-OCRv5 models on ONNX Runtime — accurate on mixed Korean/English, code, and UI text. Free, fully offline, no language pack required (models ship inside the exe).
• Many capture modes — region · full screen · click-to-capture a window (auto-detected) · repeat last region · delayed (3 / 5s).
• Capture history gallery. Browse every saved capture as thumbnails → re-drag, copy, edit, or delete (to Recycle Bin).
• Filename templates — {app}, {title}, {date}, {seq}, {w}, {h} tokens (the foreground app / window title are captured at grab time).
• Pin to keep a thumbnail (disables auto-dismiss and promotes the file out of %TEMP%).
• GIF recording · scrolling capture · clipboard-image detection.
• One dark design system across the overlay, editor, and settings — including dark DWM title bars.
• Lean tray resident. Idle memory was cut hard in 1.2.4 (single-digit working set / ~85 MB committed); the OCR engine loads lazily and releases after use, so OCR doesn't tax the idle footprint.
• Optional sharing. Imgur upload (bring your own Client-ID) from the thumbnail.
• Control it from the terminal or an AI agent. A built-in MCP server (wsnap mcp, for Claude Desktop / Claude Code) plus a wsnap <verb> CLI expose capture, OCR, color, GIF, and history through one command bus. Off by default — turn it on in Settings.

Install

Windows 10/11, x64.

Package managers

# Scoop
scoop install https://raw.githubusercontent.com/openwong2kim/wsnap/main/packaging/scoop/wsnap.json

# winget (once the manifest is accepted into winget-pkgs)
winget install openwong2kim.wsnap

Direct download — grab the latest from Releases:

• wsnap-setup-x.y.z.exe — installer (Start Menu shortcut, optional run-at-startup)
• wsnap-vx.y.z-win-x64.zip — portable single .exe, no install

wsnap is not code-signed yet, so Windows SmartScreen may show an "unknown publisher" prompt — click More info → Run anyway. (OSS code signing is wired up and pending; see SIGNING.md.)

Usage

1. Launch — no window appears, just a tray icon.
2. Press Shift+F1 (or double-click the tray icon) and drag a region.
3. The thumbnail at the bottom-right:
   • Left-click drag → hand off the file (stays available to drag again elsewhere)
   • Click → copy the file path
   • Hover buttons → Edit / Text (OCR) / ✕
   • Right-click drag (sideways) → flick it away
   • Leave it → auto-dismisses after your configured delay
4. Tray menu: Capture · OCR region · GIF record · Scrolling capture · Clear all · Settings · Quit.

Settings: save folder, hotkey rebinding, auto-dismiss delay, max thumbnails shown, run at startup, intercept Win+Shift+S, history (date folders), clipboard detection, telemetry (opt-in), upload.

Automate & control (CLI · MCP)

wsnap can be driven from a terminal or by an AI agent through one shared command bus. This is off by default — turn on External control in Settings first (with it off, the control pipe server is never even created).

CLI — wsnap <verb>:

wsnap capture --full                       # the monitor under the cursor
wsnap capture --region 0,0,800,600         # a fixed rectangle -> saved PNG
wsnap capture --window --copy              # foreground window, straight to the clipboard
wsnap ocr --last                           # OCR the most recent capture
wsnap ocr --file shot.png --lang latin     # OCR an image file
wsnap color --cursor --format hex          # pixel color under the cursor
wsnap gif --region 0,0,640,480 --duration 5
wsnap history list --limit 10

Add --json for a single machine-readable object on stdout, or --out - to stream raw PNG bytes. Exit codes: 0 ok · 1 error · 2 usage · 3 needs the running app · 4 no result / cancelled · 5 OCR unavailable. Run wsnap --help for the full list.

MCP — register the wsnap mcp server with an MCP client. For Claude Code:

claude mcp add wsnap -- wsnap mcp

For Claude Desktop, add to %APPDATA%\Claude\claude_desktop_config.json (Settings has a Copy MCP config button for this snippet):

{ "mcpServers": { "wsnap": { "command": "wsnap", "args": ["mcp"] } } }

The server exposes twelve tools: capture_region, capture_fullscreen, capture_window, capture_interactive, ocr_region, ocr_image, ocr_last_capture, pick_color, list_history, get_capture, record_gif, stop_recording.

Security. Control is local-only — a per-user named pipe plus MCP stdio, with no network listener. Every agent-initiated capture raises a visible signal (shutter flash / toast / tray badge) and is written to an audit log (%APPDATA%\wsnap\audit.log). Returning captured pixels or OCR text to the caller is a separate opt-in (Allow returning captured content, off by default), and a per-minute rate limit applies.

Build from source (Windows)

Requires the .NET 8 SDK (or 9) with the Windows Desktop workload. The project targets net8.0-windows10.0.19041.0 for WinRT projection availability.

The Korean OCR model lives in models/v5/ and is embedded into the exe at build time. To (re)download it:

pwsh -File tools\fetch-ocr-models.ps1

Run, publish a single self-contained exe, and build the installer:

dotnet run --project Wsnap.csproj          # run from source
pwsh -File publish.ps1                      # -> publish\wsnap.exe (single file)
ISCC.exe installer.iss                      # -> dist\wsnap-setup-x.y.z.exe (Inno Setup 6)

Source map

File	Role
App.cs	Entry point · tray · capture modes · action routing · single instance
Theme.cs	Shared design system (color · type · control styles · dark title bars)
Icons.cs	Vector line icons (font-independent)
ImageClipboard.cs	Multi-format image clipboard (DIB + PNG + FileDrop, with retry)
HotkeyHook.cs	Global keyboard hook (custom hotkey + Win+Shift+S toggle)
CaptureOverlay.cs	Capture overlay (freeze · dim · W×H · loupe · action toolbar · Capture/OCR/Region/ColorPick)
ScreenGrab.cs	Screen pixel grab + Bitmap→BitmapSource
CaptureStore.cs	Save-location / history policy + pin promotion
ThumbnailWindow.cs	Floating thumbnail stack (copy · save · edit · OCR · folder · share · pin · delete)
HistoryWindow.cs	Capture history gallery (thumbnail grid · drag-out · re-edit · delete)
EditorWindow.cs	Annotation editor (11 tools · redo · undoable crop · copy to clipboard)
Ocr.cs	PaddleOCR PP-OCRv5 (ONNX / RapidOcrNet) wrapper — Korean rec model, lazy load + idle release
GifRecorder.cs / GifWriter.cs	GIF recording + delay/loop encoding
ScrollCapture.cs	Scrolling capture (overlap stitching)
ClipboardWatcher.cs	Clipboard image detection
Uploader.cs	Imgur upload
Settings.cs / SettingsWindow.cs	Settings model · UI (dark cards)
AutoStart.cs / SingleInstance.cs / CrashLog.cs / Toast.cs	Tray-resident infrastructure
Control/Command.cs / CommandCatalog.cs / CommandRouter.cs	Control-layer contracts · command catalog (id ↔ MCP tool ↔ CLI) · local command bus
Control/CaptureCore.cs / ControlGate.cs / AuditLog.cs	Headless capture/OCR/color facade · consent · rate-limit gate · audit log
Control/CliRouter.cs / ConsoleBridge.cs	wsnap <verb> CLI (parsing · output contract) · WinExe console attach
Control/McpStdioServer.cs / JsonRpc.cs	wsnap mcp stdio MCP server · minimal JSON-RPC 2.0
Control/PipeServer.cs / PipeClientRouter.cs / PipeProtocol.cs	Per-user named-pipe control server · client delegation · NDJSON framing
Control/FolderWatcher.cs	Watch-folder auto-OCR
App.Control.cs	Resident host — interactive / recording delegation · external-control wiring

Good to know

• OCR: PP-OCRv5 models are embedded in the exe — no language pack needed. The first recognition is slightly slower (engine warm-up), then fast; the engine is released after a short idle. Rotated text isn't de-skewed (screenshots are assumed upright).
• Scrolling capture is best-effort — solid on text and web pages, weaker on smooth-scroll / parallax content.
• Privacy: no tracking. Telemetry is opt-in and local-log only (%APPDATA%\wsnap\wsnap.log).
• External control (CLI / MCP) stays off until you enable it, talks only over local stdio and a per-user named pipe (no network listener), and records every agent-initiated capture — each with a visible on-screen signal — to %APPDATA%\wsnap\audit.log.
• Code signing is recommended before wide distribution to avoid SmartScreen — see SIGNING.md / ROADMAP.md.

License

GPL-3.0-only. © 2026 openwong2kim and wsnap contributors.

OCR is powered by RapidOcrNet and PaddleOCR PP-OCRv5 models on ONNX Runtime. Bundled third-party components and their licenses (all permissive / GPL-3.0-compatible) are listed in THIRD-PARTY-NOTICES.md.

See ROADMAP.md for detailed status and CHANGELOG.md for release history.