cleanshotx

name	cleanshotx
description	Automates CleanShot X on macOS via its cleanshot:// URL scheme — captures screenshots (area, window, fullscreen, scrolling, previous-area, all-in-one), records the screen, OCRs text from images or screen regions, pins images on top, opens files for annotation, manages capture history, toggles desktop icons, and opens specific settings tabs. Use when the user mentions CleanShot, CleanShot X, cleanshot:// URLs, scripted macOS screenshots or screen recordings, programmatic screen OCR on a Mac, pinning a screenshot, or driving the CleanShot app from a shell, Shortcut, or agent. macOS only. Most commands need CleanShot 3.5.1+; the `action=` parameter and most spatial parameters require 4.7+; `capture-text` requires 3.8.1+ on macOS 10.15+.
license	MIT
compatibility	macOS with CleanShot X installed (https://cleanshot.com). Commands fire by opening cleanshot:// URLs (`open "cleanshot://…"`); requires an Aqua login session and CleanShot to be running or launchable. Over SSH the URL fires only when the same user has an active graphical session. Most commands require 3.5.1+; `action=` and most spatial parameters require 4.7+; `capture-text` requires 3.8.1+ on macOS 10.15+; `all-in-one` 4.2+; `open-history` 4.4+.
metadata	{"app-website":"https://cleanshot.com","api-docs":"https://cleanshot.com/docs-api","url-scheme":"cleanshot://"}

Driving CleanShot X via the cleanshot:// URL scheme

CleanShot X is a macOS screenshot and screen-recording app from MagicLasso. It exposes a URL scheme (cleanshot://command?param=value&…) that triggers any of its capture, recording, OCR, annotation, pin, history, or settings actions. Opening such a URL launches the app (if needed) and runs the action. That URL scheme is the only supported automation surface — there is no public CLI, AppleScript dictionary, or HTTP API. macOS Shortcuts.app actions internally invoke the same scheme.

The full per-command reference with every parameter and version constraint lives in references/url-scheme.md. Read it before automating any command this SKILL.md does not show in full.

Gotchas — read these before automating

• Requires an Aqua login session. open "cleanshot://…" goes through LaunchServices and the app's UI. It needs an active logged-in graphical session for the same user — over SSH it fires only when that user is also logged in locally. It will not run from a launchd Background agent or before login. Use a user-context launchd agent (LimitLoadToSessionType: Aqua), Screen Sharing, or run it as a step inside a logged-in session.
• CleanShot X must be installed and able to launch. If the app is not installed, opening the URL silently does nothing or prompts the user to choose an app. Verify with mdfind "kMDItemCFBundleIdentifier == 'pl.maketheweb.cleanshotx'", or check the common install paths: /Applications/CleanShot X.app (direct/App Store) and ~/Applications/Setapp/CleanShot X.app (Setapp).
• No return value, no completion signal. The URL fires asynchronously; the calling shell gets exit code 0 as soon as open hands off. There is no built-in way to wait for "capture finished" or to recover the resulting file path. To know what happened: inspect the configured save folder (CleanShot ▸ Settings ▸ Screenshots ▸ "Save to") or watch the clipboard.
• URL parameters must be URL-encoded. Spaces in a filepath become %20. Always quote the URL when passing it to open, and percent-encode user-supplied paths. See the helper at the bottom of this file.
• Version-gated parameters may not work on older builds. Parameters added in 4.7 (action=…, x/y/width/height/display on most commands, start=true, autoscroll=true, tab=…) are not supported on earlier versions and are typically ignored — the command still runs without the parameter. Don't rely on a version-gated parameter without first checking defaults read "/Applications/CleanShot X.app/Contents/Info" CFBundleShortVersionString.
• action is a single value, not a chain. action=annotate alone is valid; action=copy,upload is not. Pick one of copy | save | annotate | upload | pin. To compose actions (e.g. annotate then upload), drive subsequent steps from a follow-up URL or from the annotator UI.
• filepath parameter rules. When you supply a filepath to any command that takes one (, , , ), it must be an — not -relative. Expand in the shell before passing. The parameter is for / / (omitting it falls back to an interactive picker or area selection — see ) and for . Accepted formats: PNG and JPEG for screenshot commands; additionally accepts MP4.

Quick start: invoking the URL scheme

The single primitive is open with a cleanshot:// URL. From a shell:

# Launch the area-selection capture, then open the result in the annotator.
open "cleanshot://capture-area?action=annotate"

# Repeat the previous-area capture and copy the result to the clipboard.
open "cleanshot://capture-previous-area?action=copy"

# OCR text out of an existing image and copy it to the clipboard.
open "cleanshot://capture-text?filepath=/Users/me/Desktop/screenshot.png&linebreaks=true"

# Pin an image on top of all windows.
open "cleanshot://pin?filepath=/Users/me/Desktop/diagram.png"

# Open the Recording tab in CleanShot's settings.
open "cleanshot://open-settings?tab=recording"

From AppleScript / osascript:

osascript -e 'open location "cleanshot://capture-fullscreen?action=upload"'

From Python (when you need URL-encoding done for you):

import subprocess, urllib.parse, pathlib
path = pathlib.Path("~/Desktop/My Screenshot.png").expanduser().resolve()
url = "cleanshot://pin?filepath=" + urllib.parse.quote(str(path), safe="/")
subprocess.run(["open", url], check=True)

Command catalog (summary)

The 19 commands fall into seven groups. For full parameters, defaults, and per-parameter version requirements, see references/url-scheme.md.

Group	Commands
Screenshots	capture-area, capture-previous-area, capture-fullscreen, capture-window, self-timer, scrolling-capture, pin
Recording	record-screen
OCR	capture-text
Annotation	open-annotate, open-from-clipboard
All-In-One	all-in-one
Desktop icons	toggle-desktop-icons, hide-desktop-icons, show-desktop-icons
History / overlays / settings	add-quick-access-overlay, open-history, restore-recently-closed, open-settings

Common parameter shapes:

• action=copy | save | annotate | upload | pin — what to do with the result. Requires CleanShot 4.7+. Default is whatever the user has configured in CleanShot ▸ Settings ▸ Screenshots ▸ "After capture".
• x, y, width, height, display — capture region. Requires 4.7+ on most commands. Coordinates use the macOS native coordinate system: origin (0,0) is the lower-left corner of the screen, with y increasing upward (per the CleanShot docs). Units are points, not Retina pixels. display is a 1-based integer; if omitted it defaults to the display under the cursor.
• filepath — absolute path to a PNG or JPEG (MP4 also for add-quick-access-overlay). Must be URL-encoded.

Common workflows

One-shot: capture an area and upload to the cloud

open "cleanshot://capture-area?action=upload"

The user drags out a region; CleanShot uploads to the configured destination (CleanShot Cloud, S3, Dropbox, etc.) and copies the resulting URL to the clipboard. The URL is not returned to the caller — read it from the clipboard:

pbpaste   # → https://cln.sh/abcd1234

Repeat the previous capture region and save to disk

Useful for taking a sequence of "same region" shots while a UI updates:

open "cleanshot://capture-previous-area?action=save"

OCR text out of an image file

open "cleanshot://capture-text?filepath=$(python3 -c 'import urllib.parse,sys;print(urllib.parse.quote(sys.argv[1]))' "$HOME/Downloads/menu.jpg")&linebreaks=false"
sleep 1
pbpaste   # OCR'd text

linebreaks=false flattens the text into a single paragraph; true preserves line breaks. CleanShot's standard OCR flow writes the recognized text to the clipboard — read it with pbpaste (the URL itself returns nothing).

Capture a specific named window unattended

The right tool for "screenshot Ghostty's window" with no clicks is not capture-window (that command always opens an interactive picker). It is capture-area with explicit x/y/width/height (4.7+) computed from the window's frame, plus action=save. The only twist is the coordinate flip: macOS UI APIs (System Events, AppKit, Accessibility) report top-left-origin bounds, but CleanShot's URL scheme uses the lower-left origin (cs_y = display_height - top_y - height).

The snippet below uses only tools shipping in stock macOS — osascript (AppleScript and JXA), defaults, find. Do not reach for PyObjC / import Quartz / pip install pyobjc here: those are not part of a stock macOS python3, and any agent that tries them will hit ModuleNotFoundError: No module named 'Quartz'. AppleScript covers everything you need.

There are two flavours of "the right window":

• "This window" (the one the user is currently looking at, e.g. when the user types "screenshot this window" inside the agent's host terminal). Resolve the frontmost application's frontmost window — do not hard-code an app name like Ghostty, because that breaks the moment the user runs the agent inside iTerm2, Terminal.app, Alacritty, Wezterm, Kitty, or any other terminal.
• "Window of app X" (the user explicitly named the app, e.g. "screenshot Slack"). Pin the AppleScript to that app by name.

Both flavours share the rest of the pipeline (raise the window so it isn't occluded; flip y; fire the URL; poll for the file). The only difference is how APPNAME and the window-bounds query are resolved at the start.

# 1a. Resolve "this window" — the frontmost app's frontmost window. Atomic
#     in one AppleScript pass so the answer can't drift between calls.
read -r APPNAME WX WY WW WH <<<"$(osascript <<'AS'
tell application "System Events"
  set frontApp to first application process whose frontmost is true
  -- Skip menu-bar-only apps (no windows) and walk to the next visible app.
  if (count of windows of frontApp) = 0 then
    repeat with proc in (every application process whose visible is true)
      if (count of windows of proc) > 0 then set frontApp to proc
      if (count of windows of frontApp) > 0 then exit repeat
    end repeat
  end if
  set w to window 1 of frontApp
  -- Raise it so a region-capture doesn't grab whatever is currently on top.
  try
    set frontmost of frontApp to true
  end try
  try
    perform action "AXRaise" of w
  end try
  delay 0.2
  set pos to position of w
  set sz to size of w
  return (name of frontApp) & " " & (item 1 of pos as text) & " " & (item 2 of pos as text) & " " & (item 1 of sz as text) & " " & (item 2 of sz as text)
end tell
AS
)"

# 1b. Alternative: pin to a specific named app (e.g. APP="Slack").
#     Use this when the user explicitly named the target app.
#     APP="Slack"
#     read -r APPNAME WX WY WW WH <<<"$(osascript <<EOF
#     tell application "System Events" to tell process "$APP"
#       set w to window 1
#       try
#         set frontmost to true
#       end try
#       try
#         perform action "AXRaise" of w
#       end try
#       delay 0.2
#       set p to position of w
#       set s to size of w
#       return "$APP" & " " & (item 1 of p as text) & " " & (item 2 of p as text) & " " & (item 1 of s as text) & " " & (item 2 of s as text)
#     end tell
#     EOF
#     )"






SH=$(osascript -l JavaScript -e )


CSY=$(( SH - WY - WH ))


MARK=$( -t cs.marker)
open 




SAVE_DIR=$(defaults  pl.maketheweb.cleanshotx exportPath 2>/dev/null \
  ||  )


 _  1 2 3 4 5 6 7 8 9 10; 
  NEW=$(find  - f -newer  \
    \( -name  -o -name  \) - -quit 2>/dev/null)
  [ -n  ] && {  ; ; }
   0.3

Add &display=N (1-based) to the URL if you also want CleanShot to disambiguate; the capture-area math above already handles non-primary displays through JXA's NSScreen.screens lookup. CleanShot may still flash a save-confirmation overlay unless the user has set "After capture: Save" in their settings.

If you have a single primary display and want a one-liner instead of the JXA snippet, AppleScript's Finder also returns the main display height in points: osascript -e 'tell application "Finder" to get bounds of window of desktop' | awk -F', ' '{print $4}'. Same caveat — points, not pixels.

Caveat: multi-tab/multi-window terminals and "this window"

System Events models a macOS application process as having an ordered list of windows. It does not expose tabs, panes, or per-TTY window mappings. So when the agent runs inside one tab/pane of a multi-tab terminal (Ghostty, iTerm2, Wezterm, Kitty, Alacritty…) and the user has also opened other windows or tabs in that same terminal app, the recipe above resolves to the application's most-recently-active window — which may not be the agent's own host tab/window.

Tempting fixes that don't work reliably:

• OSC 0/2 title-marker injection (writing an ESC ] 0 ; MARKER BEL sequence to /dev/<user-tty>). The marker reaches the right pty, but Claude Code (and many shell prompts) re-set the window title on every render, so the marker is overwritten well under a second after you write it — System Events usually reads the post-restore title, not the marker.
• screencapture -l <windowID>. It bypasses CleanShot entirely and produces a file CleanShot's pipeline doesn't know about; only fall back to it when CleanShot is genuinely unavailable.
• Ghostty's GHOSTTY_SURFACE_ID env var. Ghostty exposes the surface id but does not (as of 1.3.x) provide a CLI/AppleScript hook to focus a surface by id, so reading the env var doesn't get you to the right window.

What to do instead:

1. Resolve the frontmost app + its frontmost window atomically (the snippet above) and capture immediately. In the common case — single-window terminal, or the user is actively staring at the agent — this is correct.
2. If the user explicitly names a target ("screenshot Slack"), use the APP=… variant in 1b so the result doesn't depend on what happens to be frontmost.
3. If the user has multiple windows of the same app and demands the agent's host window specifically, the only fully reliable path is to ask them to bring that window to the front first, or to run the agent in a single-window terminal session for the duration of the capture.

Unattended scrolling capture of a long page (4.7+)

open "cleanshot://scrolling-capture?start=true&autoscroll=true"

Without autoscroll=true, the user has to scroll manually; without start=true, they have to click Start in the overlay.

Pin an image as a floating reference

ABS_PATH=$(python3 -c 'import os,sys;print(os.path.abspath(os.path.expanduser(sys.argv[1])))' "~/Desktop/spec.png")
ENC_PATH=$(python3 -c 'import urllib.parse,sys;print(urllib.parse.quote(sys.argv[1]))' "$ABS_PATH")
open "cleanshot://pin?filepath=$ENC_PATH"

The image floats above all windows until the user closes the pin. The URL scheme has no documented unpin command — close the pin manually or via UI scripting.

Locate the file CleanShot just saved

The open "cleanshot://…" call returns immediately and prints nothing — there is no synchronous way to recover the saved file path. Three workable patterns:

1. Poll the configured save folder. Read the user's "Save to" path from pl.maketheweb.cleanshotx's exportPath default (do not read CaptureFolder — that key does not exist; defaults read will error out and your fallback to ~/Desktop will look in the wrong place for any user who has changed their CleanShot save folder), then pick the newest file written after a marker:

   SAVE_DIR=$(defaults read pl.maketheweb.cleanshotx exportPath 2>/dev/null \
     || echo "$HOME/Desktop")
   MARK=$(mktemp -t cs.marker)
   open "cleanshot://capture-area?x=0&y=0&width=800&height=600&action=save"
   # Wait briefly for CleanShot to write the file, then grab the newest one.
   for _ in 1 2 3 4 5; do
     NEW=$(find "$SAVE_DIR" -type f -newer "$MARK" \
       \( -name '*.png' -o -name '*.jpg' \) -print -quit)
     [ -n "$NEW" ] && { echo "$NEW"; break; }
     sleep 0.5
   done
   

   find … -newer works because CleanShot writes a brand-new file per capture. (annotateLastSaveURL in the same defaults domain also tracks the most recently saved capture, but it's only updated when CleanShot's annotator wrote the file — exportPath + a fresh-file scan is more reliable.)

2. Use action=copy instead of save. The image goes on the clipboard; pull bytes with pbpaste (or the AppKit NSPasteboard) and write them yourself.

3. CleanShot copies the upload URL (not the image) to the clipboard — returns it, e.g. .

Do not assume open prints the file path: it never does, regardless of action=.

Detect installed version before using 4.7-only parameters

VERSION=$(defaults read "/Applications/CleanShot X.app/Contents/Info" CFBundleShortVersionString 2>/dev/null || echo "0")
# Compare with sort -V; bail out cleanly if too old.
if [ "$(printf '%s\n4.7\n' "$VERSION" | sort -V | head -n1)" != "4.7" ]; then
  echo "CleanShot $VERSION is too old for action= / spatial parameters" >&2
  exit 1
fi

URL-encode a path safely (helper)

CleanShot accepts only percent-encoded URLs. The most reliable encoder for an arbitrary path:

encode_path() {
  python3 -c 'import sys,urllib.parse,os;print(urllib.parse.quote(os.path.abspath(os.path.expanduser(sys.argv[1])),safe="/"))' "$1"
}
open "cleanshot://open-annotate?filepath=$(encode_path "~/Pictures/My Shot.png")"

Bash-only fallback (no Python): printf '%s' "$path" | jq -sRr @uri if jq is available.

Verifying CleanShot is present and runnable

Before firing any URL, agents should confirm:

# 1. App installed? (covers /Applications and Setapp)
if [ ! -d "/Applications/CleanShot X.app" ] && [ ! -d "$HOME/Applications/Setapp/CleanShot X.app" ]; then
  # Fallback: query LaunchServices by bundle ID (also handles non-standard paths)
  if [ -z "$(mdfind "kMDItemCFBundleIdentifier == 'pl.maketheweb.cleanshotx'")" ]; then
    echo "CleanShot X not installed" >&2; exit 1
  fi
fi

# 2. URL scheme registered? (LaunchServices)
/System/Library/Frameworks/CoreServices.framework/Versions/A/Frameworks/LaunchServices.framework/Versions/A/Support/lsregister -dump 2>/dev/null \
  | grep -A1 "scheme:" | grep -q "cleanshot" \
  && echo "URL scheme registered"

# 3. Logged-in graphical session? (must be Aqua)
launchctl managername | grep -qi Aqua || echo "Warning: not an Aqua session — URL scheme may not fire"

When the URL scheme isn't enough

Things the URL scheme cannot do — escalate or document the limitation rather than inventing a command:

• Programmatically stop a recording (no stop-recording URL).
• Compose multiple post-capture actions in one call (action=copy,upload).
• Read the resulting file path or upload URL from the command — must read the clipboard or scan the configured save folder.
• Drive Quick Access overlay placement, size, or z-order beyond "added".
• Change settings programmatically (only open the settings tab).
• Capture without a user-visible UI (every command surfaces some on-screen affordance).
• Unpin a pinned image.

For workflows that need richer integration than the URL scheme exposes, fall back to macOS Shortcuts.app (CleanShot ships several Shortcut actions that wrap these URLs). Drop down to OS-level alternatives (screencapture, xcrun simctl io … screenshot, native AVFoundation recorders) only when CleanShot is genuinely unavailable. In particular, do not reach for screencapture -l <windowID> to dodge capture-window's click — cleanshot://capture-area with explicit x/y/w/h (see Capture a specific named window unattended) covers that case headlessly.

Where to look next

• references/url-scheme.md — every command, every parameter, version requirements, examples.
• Vendor docs: https://cleanshot.com/docs-api (authoritative; check for new commands when this skill seems incomplete).
• App version: defaults read "/Applications/CleanShot X.app/Contents/Info" CFBundleShortVersionString.
• App support directory (per-user CleanShot state): ~/Library/Application Support/pl.maketheweb.cleanshotx/.
• Default save folder for captured files: shown in CleanShot ▸ Settings ▸ Screenshots ▸ "Save to".