OpenScreen MCP
OpenScreen MCP: an agent records the screen, trims and zooms frames, adds captions and exports MP4 or GIF.
- 7,026 ★ on the repository
- Last commit
Broad access
Why this level
- The agent records the whole screen, microphone and system audio with the user rights.
- Third-party data can land in frames and transcripts, and posting the video to a pull request exposes it.
Install
- Works in
Manual install
claude mcp add openscreen -- "<executable>" mcpThe README command for Claude Code, where `<executable>` is the OpenScreen app executable.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent records the whole screen, microphone and system audio with the user rights.
- Third-party data can land in frames and transcripts, and posting the video to a pull request exposes it.
This is third-party code. Review the repository files before installing.
What it does
OpenScreen is a free screen recording editor. Its MCP server is started by the app itself with `<executable> mcp` over stdio: the agent launches the process, and no port, token or setting is needed. Through the server the agent records the screen, edits the timeline and exports the result. The full cycle uses `listSources`, `startRecording`, `stopRecording`, `setPointerTrack`, `exportVideo`, `exportAudio`, `getFrames` and `createProject`, plus editing tools for trims, zooms, speed and annotations. The agent requests a voice transcript with `transcribe`, and Whisper runs on your machine. The `openscreen-demo` skill sits alongside: it leads the agent from a storyboard to recording, editing, an MP4 export and attaching the video to a pull request. The same skill ships as a plugin from the repository marketplace.
Who it is for. Developers who show changes as video demos in pull requests and releases and want to hand recording and editing to an agent in Claude Code, Codex or Cursor.
Good fit when
- A pull request needs a video of the change: the agent records the app, edits it and attaches the MP4.
- A release demo is prepared with zooms on clicks and captions.
- Words are cut from a recording with a voiceover using the transcript and subtitles are turned on.
- A Playwright demo script becomes a finished edited video.
Not a fit when
- A studio-grade edited video is needed: the OpenScreen edit set is built for demos.
- The machine screen must not be recorded: on the agent request the server records the screen, microphone and system audio.
- OpenScreen is not installed: the server needs its executable.
Example request
Record how the new sign-in form works, zoom on the clicks, add captions, export an MP4 and attach it to the pull request.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
The server is started by the OpenScreen app executable, so the app must be installed. The agent starts the process itself and acts with the user rights: it can record the screen, microphone and system audio, per the documentation even without the MCP edits switch turned on. The project is under active development and the `.openscreen` format and the command line may break. Undo for agent edits works through checkpoints, the 20 most recent are kept and only until the agent session ends. The agent gets new tools only in a new session. Voice transcription is done by Whisper on your machine.
How to disable. Remove the `openscreen` entry from the client settings (`claude mcp remove openscreen` for Claude Code) and, if installed, the `openscreen@openscreen` plugin. Recordings and projects stay in the OpenScreen project list.
FAQ
How do I connect Claude Code?
Run `claude mcp add openscreen -- "<executable>" mcp`, with the path to the OpenScreen app in quotes. Paths for each system are collected in the Running the CLI section on the project site.
How do I connect Codex?
Run `codex mcp add openscreen -- "<executable>" mcp`. The Codex plugin installs with `codex plugin marketplace add getopenscreen/openscreen` and `codex plugin add openscreen@openscreen`.
What does the openscreen-demo skill do?
It installs OpenScreen with your consent, connects MCP and asks you to open a new session, where the agent sees the new tool set. Then it leads the demo from storyboard to a video in the pull request.
Is my own model needed?
Not for MCP itself. The built-in editor assistant needs your own key (Claude, OpenAI, Gemini, Mistral and others) and is off by default.
Is there a command line?
Yes, for example `openscreen record --duration 20 --project demo.openscreen --json` and `openscreen export demo.openscreen -o demo.mp4 --json`.
Discussion
No questions or reviews yet.
Related
MCP serversScreenplay films screen recordings from one prompt: the agent writes a scenario, the tool plays and edits it.
An MCP server with 59 tools for CapCut and OpenCut editing: tracks, clips, text, transitions and voiceover from Cursor and Codex.
MCP server, CLI and desktop app that cut a long video or stream replay into vertical clips with captions
ClipForge MCP
ClipForge
ClipForge MCP builds a vertical video from a topic or a product link: stock footage, Edge TTS voiceover and captions, no keys, composed with FFmpeg
SmartCut MCP
SmartCut MCP Server
MCP server that edits talking head videos inside a CapCut project: it uses auto-captions to cut silences and repeated takes.
An MCP server for video editing built into the free Concat editor: the agent cuts clips, adds titles and captions, sets keyframes and exports video
README in short
The README covers recording features, the built-in AI assistant, a command line for agents and the demo skill, with installation per platform.