ClickOCR

Point at a word on screen, press a hotkey, click — the text lands on your clipboard.

Windows has a very good OCR engine (Windows.Media.Ocr) built in. ClickOCR is a thin tray app around it that answers "what does this word say?" instead of "OCR this box":

Ctrl+Alt+O   ->  screen freezes and dims  ->  click the word  ->  clipboard

Recorded with --demo: the first click copies the whole line, Ctrl+click copies one word; the panel under the sample text is the live clipboard. Bigger MP4.

• Runs offline, no cloud, no account, no install, no admin rights.
• ~57 KB single .exe, .NET Framework 4.8 + the OCR engine that ships with Windows.
• Works on any visible text: browsers, PDFs, screenshots, images, games, videos, error dialogs.
• Knows where the word is, because it uses the OCR word bounding boxes to decide what you clicked.

Quick start

cd ClickOCR
powershell -ExecutionPolicy Bypass -File .\build.ps1      # builds bin\ClickOCR.exe
.\bin\ClickOCR.exe                                        # tray icon appears

Then press Ctrl+Alt+O, click some text, paste.

No build toolchain? Take ClickOCR.exe from the latest release.

Want to see it work first?

.\bin\ClickOCR.exe --demo

A sample window opens: a fake invoice plus a live clipboard panel, so every click shows its result without pasting anywhere. It also writes the on-screen coordinates of a few words to %APPDATA%\ClickOCR\demo-anchors.txt, which is what tests\e2e.ps1 clicks.

Using it

Action	Result
Ctrl+Alt+O, then click	the whole line / string under the pointer
Ctrl+Alt+O, then Ctrl+click	just the word under the pointer
Ctrl+Alt+O, then Shift+click	every line in the captured area
Esc or right-click	cancel the picker
Double-click the tray icon	same as the hotkey

After a grab a small bubble shows what was copied; it never steals focus, so you can keep typing in the app you came from.

The picker freezes the screen first, so nothing redraws or scrolls while you aim. It also cancels itself after 45 seconds so it can never leave your desktop dimmed.

Tray menu

• Grab text — same as the hotkey.
• Settings... — opens %APPDATA%\ClickOCR\config.ini in Notepad.
• Start with Windows — writes the usual HKCU\...\Run entry (with --silent).
• Run self-test... — checks OCR, capture, hit-testing and clipboard; opens a report.
• Open log — %APPDATA%\ClickOCR\log.txt, every grab with timings and result.
• Help / About, Exit.

Settings — %APPDATA%\ClickOCR\config.ini

Created with defaults on first run; restart the app after editing.

Hotkey=Ctrl+Alt+O       ; global shortcut
RegionWidth=1400        ; how much of the screen around the click is OCR'd (pixels)
RegionHeight=320
ClickAnchor=0.34        ; where the click sits inside that region (0 = left edge, 1 = right)
Dim=0.35                ; dimming of the frozen screen, 0 - 0.85
Language=               ; e.g. en-US, de-DE, zh-Hans-CN; empty = Windows display language
ClicklessGrab=0         ; 1 = skip the picker, OCR straight under the pointer

ClicklessGrab=1 is the "point and shoot" variant: put the pointer on a word, press the hotkey, done — no second click. (Line mode only.)

ClickAnchor matters more than it looks: text normally continues to the right of the pointer, so the default puts the click at ~1/3 of the captured width. Raise it if you often miss the left part of long lines.

Languages: only the OCR packs installed in Windows are available. See them in Settings > Time & Language > Language (the "Optical character recognition" optional feature) or in the self-test report. The current engine language is shown in Help / About.

Scripting / automation

The tray app also listens on a local named pipe (ClickOCR.Control), which is handy for AutoHotkey, Stream Decks, CI or a keyboard macro:

.\bin\ClickOCR.exe --send status                  # OK idle|picking hotkey=...
.\bin\ClickOCR.exe --send grab                    # freeze screen + show picker
.\bin\ClickOCR.exe --send pick 812 540 word       # supply click point + mode
.\bin\ClickOCR.exe --send grabcursor              # grab under the pointer (clickless)
.\bin\ClickOCR.exe --send quit

The last response is also written to %APPDATA%\ClickOCR\control-last.txt.

Tests

.\bin\ClickOCR.exe --selftest        # headless: OCR, screen capture, hit test, clipboard
powershell -File .\tests\e2e.ps1     # drives the real picker and checks the clipboard
powershell -File .\tests\clickless.ps1  # pointer + grabcursor, no click

tests\e2e.ps1 needs an idle desktop: it opens the demo window, drives the app through the control pipe, posts a real WM_LBUTTONDOWN to the picker and verifies the clipboard for line/word/block modes. Don't touch the mouse while it runs.

How it works

1. On the hotkey, the whole virtual screen is captured with CopyFromScreen (before anything is drawn), so the picker shows a frozen frame and needs no "wait for redraw".
2. The click point is mapped to a region around it (RegionWidth x RegionHeight, biased by ClickAnchor); only that region is sent to OCR — fast and accurate.
3. The bitmap is converted to a SoftwareBitmap (BGRA8) and passed to OcrEngine.RecognizeAsync. This returns lines and words, each with a bounding box.
4. The word whose box contains the click (or the nearest one within ~1.2 text heights) decides the result: the word, its whole OCR line, or every line in the region.
5. The text is written to the clipboard with retries (the clipboard is a shared resource and is occasionally locked for a few milliseconds).

The process is per-monitor DPI aware, so coordinates, OCR boxes and the capture all live in physical pixels — correct on mixed-DPI multi-monitor setups.

Limitations

• Protected content is not capturable. DRM video, some banking apps, and windows with SetWindowDisplayAffinity(WDA_EXCLUDEFROMCAPTURE) come back black — no OCR possible. Same for the Windows lock screen / secure desktop.
• OCR is not pixel-perfect: 9f2c-77b1-alpha may come back as 9f2c-77b1 -alpha, l vs 1, 0 vs O, etc. Very small text (below ~10 px) or heavy stylised fonts degrade.
• Line mode returns the OCR line as the engine grouped it. In dense tables that can be a whole row; use Ctrl+click for a single cell/word.
• The exe is unsigned, so SmartScreen or your AV may ask once. It builds locally from the source in this folder and makes no network calls.
• Hotkey registration can fail if another app owns the combination; ClickOCR then tries Ctrl+Shift+O, Ctrl+Alt+F9, Ctrl+Alt+Shift+O and tells you in a balloon.

Alternatives you can use today

If you would rather not run a custom build, these are all solid and free — but they are region selectors, not click-a-word selectors:

Tool	Shortcut	Notes
PowerToys Text Extractor	Win+Shift+T	Microsoft's own; uses the same Windows.Media.Ocr engine; drag a box, text goes to the clipboard. Closest thing to "built in".
Snipping Tool	Win+Shift+S then Text actions	Ships with Windows 11 (and recent Windows 10); rectangle snip then "Copy all text".
Text-Grab (open source)	configurable	Fullscreen grab with word/line/paragraph modes, very close in spirit to ClickOCR; .NET, MIT.
Capture2Text	Win+Q	Region capture, many engines/languages, portable.
PowerToys "Text Extractor" + OCR language packs	—	Same engine, so the same accuracy you get here.

The practical difference: with those you aim a rectangle (or a full screen) and get all the text in it, then trim it in an editor. ClickOCR turns the pointer itself into the selector, so the round trip is click → paste with no cropping step.

Layout

ClickOCR/
  build.ps1              builds bin\ClickOCR.exe (uses the in-box C# compiler + WinMD)
  bin\ClickOCR.exe       the app
  src/                   C# sources (C# 5, .NET Framework 4.8, WinForms + Windows.Media.Ocr)
  tests/e2e.ps1          end-to-end test against the real picker
  docs/demo.gif          the recording above (docs/demo.mp4 = same thing, H.264)

License

MIT — see LICENSE. Do whatever you like with it, no attribution needed beyond keeping the copyright notice.