Your AI agent can make demo videos now

Two steps, once

Command-line control is off by default. Enable it once, grant Screen Recording once, and every script or agent on your Mac can use it from then on. The app binary doubles as the CLI — there is nothing extra to install.

"/Applications/SmoothCapture.app/Contents/MacOS/SmoothCapture" cli enable

# Optional: put a short "smoothcapture" command on your PATH
"/Applications/SmoothCapture.app/Contents/MacOS/SmoothCapture" cli install-symlink

# Recommended for scripted demos: real mouse clicks for smart zoom
brew install cliclick

If the app is not running, any CLI command launches it in the background automatically.

From nothing to a polished MP4

The whole happy path is a handful of commands: check the connection, pick a capture source, record, generate smart zooms from your clicks, and export. Paste this into a terminal or hand it to your agent.

SC="/Applications/SmoothCapture.app/Contents/MacOS/SmoothCapture"

# 1. Sanity check — auto-launches the app in the background
"$SC" cli ping

# 2. List displays, windows, cameras & microphones; note an id
"$SC" cli devices --json

# 3. Start recording a display (or --window <id>, or --area x,y,w,h)
"$SC" cli record start --display 3

#    …demo your product using REAL mouse clicks (e.g. cliclick c:400,300);
#    each click becomes a zoom anchor. Pause ~1s on important UI
#    before and after clicking for the best framing…

# 4. Stop — returns the .smoothcapture project bundle path
BUNDLE=$("$SC" cli record stop --json | jq -r '.data.bundlePath')

# 5. Smart zoom — cinematic zooms generated from your clicks
"$SC" cli project autozoom --path "$BUNDLE"

# 6. Export — same Metal pipeline as the editor
"$SC" cli export --path "$BUNDLE" --output demo.mp4 --preset 1080p30

• Clicks drive the cinematography — each mouse click becomes a smart zoom anchor
• Export presets from 1080p30 up to 4K60, plus H.264, HEVC, ProRes, and GIF output
• Apply a saved editor preset headlessly for a consistent branded look
• Fine-grained edits — backgrounds, trims, aspect ratio — are plain JSON in the project bundle
• A bad take costs nothing: record cancel discards it instantly

Marketing-ready screenshots, one command

The same CLI takes stills. Raw captures come out at native retina resolution with the cursor hidden; add --pretty and the shot gets SmoothCapture's signature treatment — background, padding, rounded corners, drop shadow — matching the video editor's look.

# Raw retina capture of a window (tight bounds, transparent corners)
"$SC" cli screenshot --window 5533 --output shot.png

# Marketing-ready: background + padding + rounded corners + shadow
"$SC" cli screenshot --window 5533 --pretty --output hero.png

# Pick any bundled background, a solid color, or your own image
"$SC" cli backgrounds --json
"$SC" cli screenshot --window 5533 --background "Zen/1" --padding 0.85 \
  --output hero.jpg

Taken headlessly with cli screenshot --window --pretty — no editing afterwards.

Teach your agent in one command

A ready-made Claude Code skill packages the whole workflow — every command, the JSON output contract, error handling, and the demo-video recipe — so your agent knows when and how to drive SmoothCapture without you explaining anything. Install it once:

BASE=https://www.smoothcapture.app/skills/smoothcapture-cli
mkdir -p ~/.claude/skills/smoothcapture-cli/references
curl -fsSL "$BASE/SKILL.md" -o ~/.claude/skills/smoothcapture-cli/SKILL.md
for doc in cli-reference project-json-reference demo-video-workflow; do
  curl -fsSL "$BASE/references/$doc.md" \
    -o ~/.claude/skills/smoothcapture-cli/references/$doc.md
done

Claude Code discovers the skill automatically. From then on, a plain-English request is all it takes:

> Build the settings page, then record a short demo of it with
> SmoothCapture — smart zoom on — and attach settings-demo.mp4 to the PR.

Everything the agent sees is machine-readable: each command returns a single JSON line on stdout, so results feed straight into the next step.

$ smoothcapture cli record stop --json
{"type":"result","data":{"bundlePath":"…/demo.smoothcapture","duration":12.4}}

$ smoothcapture cli project autozoom --path demo.smoothcapture --json
{"type":"result","data":{"zoomEffectsGenerated":4,"totalZoomEffects":4}}

$ smoothcapture cli export --path demo.smoothcapture --output demo.mp4 --json
{"type":"result","data":{"outputPath":"…/demo.mp4","fileSize":645628,"duration":12.4}}

The docs behind the skill

Not using Claude Code? The docs are plain markdown — point Cursor, Codex, or any other agent at the same URLs.

What people automate with it

Anything that can run a shell command can now produce a finished demo video. With the Claude Code skill installed, agents know the workflow out of the box.

Good to know