This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Most screenshots today don't go to a photo album — they go to Claude/ChatGPT in a console, or to a chat, or you need the text inside them. Shotik is built for that world:
AI-first. One keypress puts a shot straight into Claude Code, correctly. A built-in MCP server lets Claude take screenshots itself.
Translate anything you can see. Select text in any app — or even text baked into a Telegram post, a Steam page, an image or a video — and get an instant translation. No copy-paste, no browser tab.
Speed without dialogs. Capture → it's already in the clipboard, already on disk, ready to paste.
Lightweight & honest. One native executable and nothing else. No telemetry, no account, no cloud. About 6 MB of memory and no measurable CPU in the tray (Electron → Rust).
Think ShareX capture + Flameshot annotations + a DeepL-style translator + Claude, in one small MIT app.
Related MCP server: win-capture-mcp
Shotik started as an Electron app. In 2.0 it was rewritten from scratch in Rust: every feature stayed, it reads the same settings and history, and it installs over the old version in place.
Electron 1.3.1 | Rust 2.1 | |
Installer | 71 MB | 7 MB |
On disk | 269 MB, 80 files | 28 MB, 3 files |
Processes in the tray | 6 | 1 |
Threads in the tray | 261 | 10 |
Memory in the tray | 288 MB | 5.8 MB |
Both idle in the tray at the same moment, private memory of each process:
Electron 1.3.1 Rust 2.1.1
main 67 MB 47 threads main 5.8 MB 10 threads
GPU 100 MB 120 threads
utility 69 MB 53 threads
renderer 26 MB 12 threads
utility 14 MB 20 threads
utility 12 MB 9 threads
6 processes 288 MB 261 threads 1 process 5.8 MB 10 threadsWhy memory has more than one number, why Electron was so heavy, what got worse and how it was measured: the full comparison.
Shotik tells you when a new version is out and installs it on your click, in a few seconds, keeping your settings and history. It waits while you are recording, stays quiet during a fullscreen game, and checks a signature before running anything. How it works.
Just select text — a translate bubble pops up next to it; click, and the translation appears. It works three ways, cleanest first:
UI Automation — in browsers, editors, PDFs, chat inputs: read instantly, never touches your clipboard.
On-screen OCR — in apps that draw their own text (Telegram posts, Steam, images, video frames): Shotik finds the selection highlight and reads the pixels you selected. Still no clipboard.
Ctrl+Alt+T — a manual shortcut that translates the current selection anywhere.
Prefer typing? The Text tab is a built-in translator — paste text, pick a language, done.
Translation is free out of the box (no key) or via your own DeepL key. A genuine replacement for the DeepL desktop app — and it reaches text DeepL can't (custom-drawn apps, images, video).
Copy puts the picture in the clipboard — paste into a messenger or an editor and you get the image, never a file path. The Claude action (A in the overlay) copies image and path together: Ctrl+V into Claude Code or a terminal pastes the path, into an image app — the picture.
Connect once:
claude mcp add shotik --transport http http://127.0.0.1:7464/mcpand Claude gets these tools:
Tool | What it does |
| look at the screen right now (pass |
| shoot one window by title or id — works even when it's covered by other windows or on another monitor |
| enumerate open windows: titles, apps, bounds, which display |
| open the overlay so you point at the area |
| grab your latest shot |
| capture a rectangle anywhere on the virtual desktop — global coordinates, any monitor |
| history and monitors |
| find your calls and screen recordings by path (kind, duration, size) — to transcribe or analyze |
Say "look at my screen" — and Claude sees it. Say "look here" — and just draw a box with your mouse. Say "show me the Zarya window" — and Claude reads that window's own surface, no matter what's on top of it. Say "go through my last call" — and Claude gets the path to the .m4a you just recorded.
Select text in any app and a translate bubble appears; click it for the translation. Reads the selection through Windows UI Automation (no clipboard) — and via on-screen OCR with highlight detection in Telegram, Steam, images and video. Toggle in Settings → Text & translation; Ctrl+Alt+T and the Text tab work too.
Ctrl+Shift+PrtSc)Select an area and the recognized words become selectable, right on the frozen screen — like macOS Live Text. Drag to grab words, Enter to copy, or hit Translate. Offline OCR (your Windows languages). Better than PowerToys Text Extractor — that one just dumps everything into the clipboard.
Freeze-frame, magnifier with a pixel-perfect HEX picker, boxes, arrows, pen, highlighter,
pixelate your secrets, text, numbered markers — all inside the overlay, before saving.
Hovering highlights the window under the cursor; click captures it (hold Alt to disable snapping,
Shift+click for the whole screen).
Shift+PrtSc)Same region, fresh pixels, instant, no overlay. Perfect for iteration loops:
fix the CSS → Shift+PrtSc → Ctrl+V to Claude → "see it now?". Survives restarts.
Ctrl+Alt+PrtSc)Select a region (or a window, or the whole screen) and record it to MP4 (H.264, hardware-encoded), with a floating controls bar (timer, pause, stop) and a live red border. Then export any recording to GIF in one click. Optional system-audio and microphone capture, 15–60 fps and a 3-2-1 countdown.
Enable Microphone in Settings → Screen recording and select an input device (system default unless changed). Record your voice alone or mix it with system audio into a single video audio track. The microphone is released when recording stops; unavailable devices and permission failures produce a visible error. GIF exports contain images only, without audio.
Recordings are saved with their length and a seek index, so the timeline works in any player — Windows Media Player, VLC, a browser — even for hour-long files, without re-encoding.
Ctrl+Alt+Shift+R)Record sound only: the other side of a call (system audio) and your voice (microphone), mixed
into one .m4a (AAC) that plays on any phone or messenger — about 1 MB per minute instead of
gigabytes of video. No region to select: a small bar with a timer and a live level meter per
input sits above the taskbar, so an hour of silence is caught in the first seconds. Inputs are in
Settings → Audio recording. Ready for transcription as is: ffmpeg -i call.m4a -ac 1 -ar 16000 call.wav.
Ask the other participants before recording a call.
The installer adds Start Menu commands that show up in PowerToys Command Palette and Windows search:
Area Screenshot, Full Screen, Repeat Last Area, Translate Selection, Record Audio.
CLI too: Shotik.exe --capture region|full|repeat|text|record|audio|translate.
Pin a shot on top of every window exactly where you cut it (wheel = zoom, right-click = menu)
Color picker — C in the overlay copies the pixel HEX
History gallery, toasts with previews, multi-monitor, HiDPI
Native look: light/dark theme and accent color come from the OS and switch live
English & Russian UI (follows the system language, switchable in Settings)
Action | Key |
Capture area |
|
Full screen (monitor under cursor) |
|
Repeat last area |
|
Extract text (Live Text) |
|
Record region → video / GIF |
|
Record audio (call, voice note) |
|
Translate selected text |
|
In the overlay: Enter/double-click — copy · Ctrl+S — save as · P — pin · T — OCR ·
A — copy for Claude · F — whole screen · 1..0 — tools · [ ] — stroke width ·
Ctrl+Z — undo · arrows — nudge selection · Esc — cancel.
Prebuilt: Releases →
Shotik-Setup-x.x.x.exe (installer with shortcuts, updates itself) or Shotik-x.x.x-portable.exe (single file, no install, its update button opens the release page).
From source (CI builds with Rust 1.97):
git clone https://github.com/gorka2354/shotik && cd shotik
cargo build --release -p shotik
target/release/shotik.exeWindows 11. Windows 10 (2004 or newer) should work, the build targets it, but it has not been run there yet. The app lives in the tray; quit via the tray menu. Versions before 2.1 cannot update themselves: install over them by hand, and your settings, history, autostart and shortcuts stay.
macOS: since 2.0 Shotik is Windows-only for now. The last macOS build is the .dmg in
v1.3.1.
Shotik is meant to disappear into the tray and cost nothing while it's there. Every number in this README comes from one script, bench/run.js, with the raw results in bench/results. The full table is in docs/PERF.md.
No telemetry, no account. Two things reach the network: the translation call you trigger (free service or your DeepL key), and a once-a-day look at GitHub for a newer version, which you can turn off in Settings. Screenshots stay on your disk.
Updates in one click, and only on your click. Updates are signed with a key that is not on GitHub, so taking over the repository is not enough to reach your machine. Details: docs/rust/UPDATES.md.
One executable, no runtime. The overlay stays out of memory until you need it: after you close it, the app is back to a few megabytes.
The select-to-translate watcher runs inside the app and uses Windows UI Automation to read whether text is selected. It never logs keystrokes and never touches the clipboard. It runs only while "Translate bubble on selection" is on, and you can turn it off in Settings.
Tests never touch your desktop: windows are placed off-screen, the desktop is replaced by a fixture,
global hotkeys aren't registered, and everything is driven over HTTP /test/* endpoints.
cargo test --workspace # unit and integration tests of every crate
cargo build --release -p shotik
npm test # contract tests over HTTP against the built binary (needs Node.js and ffmpeg)
node bench/run.js --impl rust --exe target/release/shotik.exe --label mine # the numbers above
node bench/charts.js # redraws the charts from bench/results
SHOTIK_BIN=target/release/shotik.exe node docs/make-screenshots.js # retakes the app screenshots, in ghost mode
python docs/promo/make_promo.py target/release/shotik.exe # re-shoots the film at the top, in ghost modeThe suite covers the parts that are easy to get subtly wrong: OCR faithfulness (small text is upscaled; Latin vs Cyrillic engines are disambiguated), selection-highlight detection, popup dismiss races, real encoded recordings decoded and measured, and the translate flow end-to-end. See AGENTS.md for the rules and docs/rust/ for the design.
crates/
shotik-core/ pure logic and ports: settings, history, translation, OCR selection, i18n
shotik-win/ Windows adapters: capture, OCR (WinRT), UI Automation, hotkeys, tray, clipboard
shotik-recorder/ screen and audio recording: Graphics Capture, Media Foundation, WASAPI
shotik-mcp/ the MCP server
shotik-ui/ windows in Slint: overlay, main window, toast, pin, translate popup, recording bar
shotik-app/ the application: triggers, overlay results, selection flow, tray menu
shotik/ the binary, wires everything togetherMIT. The Electron version it replaced stays readable at tag
electron-final.
The windows are drawn with Slint, used under its royalty-free license.
This server cannot be deployed
Connect ChatGPT, Claude, Cursor or VS Code to your Windows PC: files, terminal, browser, apps.
Hosted Screenshots MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Web scraping, browser automation, screenshots and AI data extraction for AI agents via MCP.
Grabbit gives AI agents eyes on the web through a hosted MCP server. Send a public URL and get a pixel-perfect hosted image back, without maintaining Chromium, Playwright, or a browser fleet. Capture a full page, exact viewport, or single CSS selector as PNG, JPEG, or WebP. Grabbit handles cookie and consent banners, waits for JavaScript-heavy pages, blocks private and internal URLs, supports safe retries with idempotency keys, and delivers async results through signed webhooks. Completed captures include a CDN URL. Connect with OAuth 2.1 or an API key. Grabbit works with Claude, Cursor, Codex, and any MCP client. Live captures cost $0.002 each. The $50 annual plan includes 25,000 prepaid credits that never reset or expire. Free test keys return placeholder images, so you can wire up the integration before paying. Home: https://grabbit.live Docs: https://grabbit.live/screenshot-api Built by BrainGrid.
