
OpenScreen MCP Server: Setup for Claude, Cursor and Other Apps
Lets Claude, Cursor and other assistants record your screen, edit the video and export an MP4 or GIF with the OpenScreen app.
OpenScreen is an open-source screen recorder and video editor for macOS, Windows and Linux. Its MCP server is the app's own program started with the mcp subcommand, so you need the app installed. The assistant can then record a window, cut and zoom the take, add captions and annotations, and export an MP4 or GIF.
It can record your screen and, if you ask, your microphone and webcam, so use it on a clean desktop. Recording on macOS needs Screen Recording and Accessibility permission. The project says its file format can still change between releases.
Good for
- Developers who want a demo video of a change added to a pull request
- Product teams who make launch demos with zooms, callouts and captions
- Anyone who wants the assistant to cut silences and fillers from a recording
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install OpenScreen MCP
Claude Code
- 1
Copy this prompt.
Add the OpenScreen MCP server to Claude Code. Run this command: claude mcp add openscreen -- "/Applications/Openscreen.app/Contents/MacOS/Openscreen" mcp Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires OpenScreen desktop app.
Read-only mode: There is no read-only mode. Do not approve the tools that start recording, export files or restore a checkpoint unless you want that.
Install the OpenScreen desktop app first. The server is the app's own program started with the mcp subcommand, so the app must stay installed. The path shown is the macOS one. On Windows use Openscreen.exe from the Windows installer folder, for example C:\Program Files\Openscreen\Openscreen.exe. The Microsoft Store build has no fixed path, so use the Windows installer. On Linux the .deb, .rpm and pacman packages put an openscreen command on your PATH, so use openscreen as the command, or give the full path to the AppImage. The README documents the Claude Code command; the other tabs use the same program and subcommand in each app's own format. For recording, macOS needs Screen Recording and Accessibility permission.
From the project README. Source.
Tools
List what can be recorded
Lists the displays, windows, microphones and cameras OpenScreen can record, so the assistant can pick one.
Try: “What windows can you record right now?”
Read ·
listSourcesStart recording a window or display
Starts recording one window (preferred) or one display, with the pointer saved as data so the export draws a clean cursor and plans zooms from clicks. It can add a microphone or webcam.
Try: “Record my browser window.”
Write ·
startRecordingStop recording and make a project
Stops the recording and saves it as an OpenScreen project with the given title.
Try: “Stop recording and call it Signup demo.”
Write ·
stopRecordingGive the cursor path of a scripted recording
For recordings driven by a script rather than a real mouse, you pass the pointer's path so OpenScreen draws the cursor and zooms on its clicks.
Try: “Add the cursor path from my Playwright script.”
Write ·
setPointerTrackRender the project to MP4 or GIF
Renders the project to an MP4 or a GIF with its background, cursor, zooms, cuts, speed changes and annotations, and returns the file path.
Try: “Export the project as an MP4.”
Write ·
exportVideoLook at still frames of a video
Returns still images of a video or project at the moments you pick, so the assistant can check each zoom and callout.
Try: “Show me the frames where each zoom happens.”
Files ·
getFramesCreate a new project
Creates a project, empty or from a video file laid out as one clip, with zooms planned from the clicks when the video has them.
Try: “Create a project from /Users/me/demo.mp4.”
Write ·
createProjectList your projects
Lists the OpenScreen projects the assistant can work on.
Try: “Which OpenScreen projects do I have?”
Read ·
listProjectsRename a project
Changes the project's title.
Try: “Rename the project to Launch demo.”
Write ·
setProjectInfoAdd a video or audio file to a project
Adds a video or audio file from an absolute path on your computer to the project's media.
Try: “Import /Users/me/music.mp3 into the project.”
Write ·
importMediaRemove a file from a project
Removes an asset from the project's media.
Try: “Remove the unused music file from the project.”
Delete ·
removeMediaTranscribe the speech
Transcribes the recording's speech so the transcript and captions can be edited. Only present when the server supports it.
Try: “Transcribe the recording.”
Write ·
transcribeSave a restore point
Saves the project's current state and returns a checkpoint id. It changes nothing.
Try: “Save a checkpoint before you start editing.”
Read ·
createCheckpointGo back to a restore point
Puts the project back exactly as it was at the checkpoint, discarding every edit made since, including yours.
Try: “Restore the checkpoint.”
Write ·
restoreCheckpointRead the whole project
Reads a compact snapshot of the project: media, clips, trims, and the zoom, speed, annotation, layout, audio and effect regions.
Try: “What is in this project right now?”
Read ·
getCurrentDocumentRead the transcript
Reads the transcript segments, speech and silence, with times and text.
Try: “Show the transcript.”
Read ·
getTranscriptRead the cursor path
Reads where the pointer was over time in a recording.
Try: “Where did my cursor go in the first 10 seconds?”
Read ·
getCursorTrackRead the transcript word by word
Reads the transcript one word at a time, with ids that the word tools use.
Try: “List the words with their ids.”
Read ·
getTranscriptWordsFix a misheard word
Corrects one word in the transcript. Captions follow it, but the film and audio are untouched.
Try: “Change the word 'cube control' to 'kubectl'.”
Write ·
setWordTextCut spoken words out of the film
Cuts the chosen words, such as fillers, a repeated phrase or a false start, out of the film.
Try: “Cut all the 'um's.”
Delete ·
removeWordsCut a span out of a clip
Adds one trim range, a span that will not play or export, without splitting the clip.
Try: “Cut from 12 to 15 seconds.”
Delete ·
addTrimCut many spans in one call
Adds many trim ranges at once, for example to remove every silence.
Try: “Remove all the silences longer than a second.”
Delete ·
addTrimsChange a cut
Moves or resizes an existing trim range.
Try: “Make the first cut 2 seconds longer.”
Write ·
setTrimUndo a cut
Deletes a trim range so that span plays and exports again.
Try: “Undo the cut at 12 seconds.”
Delete ·
removeTrimShorten a clip's start or end
Sets a clip's in and out points, and optionally its crop. Other clips close the gap.
Try: “Trim the first 3 seconds off the clip.”
Write ·
setClipRangeReorder a clip
Moves a clip so it plays before another clip, or last.
Try: “Move the intro clip to the start.”
Write ·
moveClipAdd a clip to the timeline
Places a clip of a video asset on the timeline, before another clip or last.
Try: “Add the second take at the end.”
Write ·
insertClipCopy a clip
Copies a clip and its trims, zooms, speed regions and annotations, placed right after it.
Try: “Duplicate the first clip.”
Write ·
duplicateClipDelete a clip
Deletes a clip from the timeline. Other clips close the gap and effects anchored to it are dropped.
Try: “Remove the last clip.”
Delete ·
removeClipReplace the timeline with kept intervals
Replaces the whole timeline with the intervals of the source you want to keep. Everything else becomes a cut.
Try: “Keep only 0 to 30 seconds and 60 to 90 seconds.”
Delete ·
replaceTimelineAdd a zoom-in
Adds a zoom-in over a span of the edited timeline, with a depth from 1 to 6.
Try: “Zoom in on the button from 5 to 8 seconds.”
Write ·
addZoomAdd many zooms in one call
Adds many zooms at once, for a planned smart-zoom pass.
Try: “Add zooms for each click.”
Write ·
addZoomsPlan zooms from your clicks
Plans zooms from the clicks recorded with each take, as the editor's Automatic zooms does.
Try: “Add automatic zooms.”
Write ·
addAutoZoomsChange a zoom
Moves, resizes or restyles an existing zoom.
Try: “Make the zoom at 5 seconds deeper.”
Write ·
setZoomSpeed up or slow down a span
Adds a speed region between 0.25x and 16x over part of the timeline.
Try: “Speed through the loading screen at 4x.”
Write ·
addSpeedChange a speed region
Moves, resizes or changes the multiplier of a speed region.
Try: “Make that speed region 8x.”
Write ·
setSpeedAdd a text, arrow or blur
Adds a text callout, an arrow, or a blur mask over part of the timeline.
Try: “Blur the email address from 10 to 20 seconds.”
Write ·
addAnnotationChange an annotation
Changes an annotation's timing, position, size, text or look.
Try: “Make the callout text bigger.”
Write ·
setAnnotationSet the video shape
Sets the exported shape, such as 16:9, 9:16, 1:1 or 4:5.
Try: “Make it vertical for a story.”
Write ·
setAspectRatioChange the project's look
Changes the background, padding, shadow and other look settings of the project.
Try: “Use a dark gradient background.”
Write ·
setEditorSettingsAdd a visual effect
Writes a visual effect as a shader and places it over part of the timeline.
Try: “Add a soft vignette.”
Write ·
addEffectChange an effect region
Moves, resizes or re-tunes an effect region.
Try: “Make the vignette stronger.”
Write ·
setEffectChange an effect itself
Changes an effect's name, code or parameters, and so every region that uses it. New code must compile first.
Try: “Change the vignette color.”
Write ·
setEffectDefinitionRead one effect
Reads one effect in full: its code, parameters and where it is used.
Try: “Show the vignette effect's code.”
Read ·
getEffectAdd a voiceover or music
Lays an already imported audio file over the recording for a span of the timeline.
Try: “Put the music under the whole video.”
Write ·
addAudioChange an audio track
Moves, re-levels, mutes, loops or fades an existing audio track.
Try: “Lower the music by 12 dB under the narration.”
Write ·
setAudioDelete a zoom, speed, annotation or other region
Deletes a zoom, speed region, annotation, layout change, audio or effect by id.
Try: “Remove the second annotation.”
Delete ·
removeModifierRead caption settings
Reads whether captions show, how they look and where they sit.
Try: “Are the captions on?”
Read ·
getCaptionsStyle and place captions
Turns captions on or off and sets their look and position.
Try: “Turn on captions with a dark plate.”
Write ·
setCaptionSettingsRead what to translate
Reads the transcript in phrase units, with ids, for translating into a caption language.
Try: “Get the phrases to translate into Spanish.”
Read ·
getCaptionTranslationUnitsStore a caption translation
Stores the assistant's translation of caption phrases in a language layer.
Try: “Save the Spanish captions.”
Write ·
putCaptionTranslationDelete a caption language
Deletes one caption language layer. The transcript is untouched.
Try: “Remove the Spanish captions.”
Delete ·
removeCaptionTranslationTeach it names and terms
Sets the names and technical terms the speech recognizer is primed with, so it spells them right.
Try: “Add OpenScreen and Kubernetes to the vocabulary.”
Write ·
setVocabulary
Read from the source code
Permissions
high risk · 5 things to watch
High risk
The assistant can record your screen and edit and export videos, reading and writing files on your computer. Keep private windows closed while it records, and have it save a checkpoint before it edits.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It can delete clips, media entries and captions, and restoring a checkpoint discards your edits
Tools cut spans, delete clips, remove media entries from a project and delete caption languages. Restoring a checkpoint puts the project back exactly as it was and discards every edit since, including yours. Removing media leaves the file on disk. Approve those calls yourself and keep a copy of important projects. github.com
It can record your screen, microphone and webcam
The recording tools capture a window or display, and can add a microphone or camera. Close private windows first, and check the output before sharing. github.com
It reads and writes files you point it at
Import and create take absolute file paths, and export writes MP4, GIF, WAV or M4A files. Check the paths before approving. github.com
Project files may change between releases
The OpenScreen docs warn that the project format can change in breaking ways. Keep a copy of important projects. getopenscreen.com
First transcription downloads a speech model
The transcribe tool runs Whisper on your computer. The first run downloads the speech model, about 264 MB according to the CLI docs, so allow for a download and a wait on long recordings. getopenscreen.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Turning on read-only | No read-only mode is documented. |
|---|---|
| Talks to | A one-time Whisper speech model download the first time you transcribe |
| Sign-in and keys | No account or key is needed. |
| Scopes |
|
Questions
Is the OpenScreen MCP server official?
Yes. It is built into the OpenScreen app and ships from its own repository. It is not a separate package.
Do I need an API key?
No. It needs no account or key. You need the OpenScreen desktop app installed.
Which systems does it work on?
OpenScreen runs on macOS 13 or later, Windows 10 (build 18362) or later, and Linux. On Linux, native capture needs xdg-desktop-portal and PipeWire. The Windows Microsoft Store build has no fixed path for the app, so use the Windows installer for the MCP command.
What can it record?
One window or one display, with the pointer saved as data. It can add a microphone and a webcam. The tool that lists sources shows what is available.
Can it make a demo video for a pull request?
Yes. The project ships a skill for this. It records the app, edits the take, exports an MP4, checks it and attaches it to the pull request. You install the skill with the command in the README.
Can I undo what the assistant did?
Partly. It can save a checkpoint before edits and restore it afterwards, but restoring discards every edit made since, including your own. On the project open in the editor, the restore is one undo step.
Is it free?
Yes. OpenScreen is open source under the MIT license.
Information
updated Nov 2, 2026
- Publisher
- Official
- Transport
- stdio
- Auth
- None
- Version
- 2.1.1
- License
- MIT
- Last commit
- Oct 31, 2026
- Open issues
- 42
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 4Hide
- github.com · github.com
- getopenscreen.com · getopenscreen.com
- github.com · github.com
- github.com · github.com