For AI agents
Drive Wispframe from your agent.
Plug Wispframe into Claude, Cursor, or any MCP-compatible agent and ask for recordings in plain language — "record my screen for 10 seconds", "remove the background of clip.mp4". Nothing to install: the running Wispframe app is the MCP server — copy the config from its ⌘ CLI & MCP panel.
Open the app
There is nothing to install: the running Wispframe app IS the MCP server. Click ⌘ CLI & MCP in the app header and copy the config it shows — it has your live port and token filled in.
No npm, no Node, no separate process. The app holds the screen/camera permissions, so your agent needs none — and every capture asks you for approval inside the app first.
The panel asks you to pick an Agent working folder the first time you open it, and writes that path into the config you copy. That is the one thing worth getting right up front: the App Store build is sandboxed, so it may only read and write inside its own container and the folder you choose here. An agent handed any other path — including ~/Movies — is refused by macOS, not by Wispframe. Your agent is told the folder in the handshake, so it plans around it instead of discovering it from a failure.
Add to your agent
Pick your agent (top-right), then drop this in:
claude mcp add --transport http wispframe http://127.0.0.1:8763/mcp \
--header "Authorization: Bearer <token from the app>"{
"mcpServers": {
"wispframe": {
"type": "http",
"url": "http://127.0.0.1:8763/mcp",
"headers": { "Authorization": "Bearer <token from the app>" }
}
}
}Just ask
Restart the agent, then talk to it:
“Record my screen for 10 seconds.”
“Take a full-screen screenshot.”
“Remove the background of cam.mp4.”
“Turn demo.mp4 into a GIF.”
What your agent can do
Every tool runs on the same zero-copy pipeline and returns the output file path back to the agent.
list_sources
Enumerate displays, windows & apps
list_cameras
Enumerate cameras, plus a wired iPhone's screen
detect_faces
Where the faces are, frame by frame, as JSON — on device
detect_text
Where text is on screen, frame by frame — find it before you blur it
detect_silence
The silent stretches, and the spans worth keeping
detect_cuts
Where the picture changes abruptly — the shot boundaries
record_screen
Record a display / window / app to MP4, with a cursor track for Studio zoom
record_camera
Camera alone, or a wired iPhone's screen at native resolution
record_pip
Screen with the webcam composited in as a bubble
record_all
Every connected display at once, each to its own file
record_live_grid
2–9 live sources into ONE file as they capture
screenshot
Full / window / region PNG
scroll_capture
Stitch a long page into one PNG — refused on the App Store build (needs Accessibility)
pick_color
Sample a pixel's HEX
remove_background
Green-screen a clip, on-device
compose
Overlay PiP, redact/blur faces, burn-in text
compose_grid
Combine 2–9 videos — layouts, keyframes, shapes
run_settings
Execute a saved app session (settings JSON)
studio
Framed backdrop, rounded corners, shadow
trim
Keep / cut time ranges
speed
Global or per-segment speed
to_gif
Clip → animated GIF
annotate_video
Arrows, boxes, text, blur — timed
subtitle_transcribe
On-device Whisper → word-timed JSON
subtitle_burn
Burn styled captions (22 presets)
subtitle_export
Convert captions to SRT / VTT
enhance
2×/4× on-device super-resolution
burn_timer
Elapsed-time clock burned into a video
annotate_image
Arrows, boxes, text on a still PNG
For MCP directories
Listing Wispframe somewhere? Copy this rather than paraphrasing — it keeps every index saying the same thing.
One line
Screen recording, editing and screenshots on macOS, driven by your agent — 29 tools over MCP.
Description
Wispframe is a Mac screen recorder that captures 4K at 60fps for under 1% CPU, and exposes its whole toolkit over MCP: record a display, window or app, take screenshots, then trim, speed, caption, upscale, remove backgrounds, compose and export GIFs. The running app IS the server — the agent needs no permissions of its own, and every capture asks the user first. Nothing is uploaded; the models run on-device.
Requirements
macOS 14+, Apple silicon. Free from the Mac App Store; the MCP server is part of the app, not a separate install.
Every tool runs inside the Wispframe app, which already holds the Screen Recording permission — there is no separate process and nothing leaves your Mac. These tools cover the common jobs; anything beyond them (keyframed layouts, multi-display capture, live grids) is available through the CLI. See the CLI reference for the full command surface.