Allows AI clients to edit video projects by cutting from transcripts, adding captions, and applying motion, transitions, and effects.
Also known as cartesiancs/cartcut, @cartesiancs/cartcut-mcp
- Stars
- 832
- Tools
- —
- Downloads, 30 days
- 194
- Latest version
- 0.1.0
- Last commit
- Nov 9, 2026
- License
- MIT
What is CartCut MCP?
CartCut allows AI clients to edit video projects by cutting from transcripts, adding captions, and applying motion, transitions, and effects.
Summary generated from the project's README
About CartCut MCP server
CartCut connects AI clients like Claude, Cursor, or Codex to the CartCut video editor running locally on your computer. This bridge allows an AI to manipulate an active project by performing tasks such as cutting video from a transcript, adding captions, and applying motion, transitions, or effects. Because the server operates via stdio and communicates with a local address, your data stays on your machine.
Setup requires Node.js 18 or later and the CartCut application must be open. The connection uses a token found in the application settings, which the server can read automatically from local config files on macOS, Windows, or Linux. Users can install the server using npx commands for Claude Code or by adding a JSON configuration for desktop clients. For advanced editing capabilities, users can also install the specific cartcut-editing plugin to give the AI specialized skills.
What you can do with it
- 1Edit video via transcriptsAsk the AI to perform cuts in the video timeline based on the text of a transcript.
- 2Apply visual enhancementsInstruct the AI to add motion, transitions, and various effects to your current project.
- 3Generate automated captionsHave the AI client add text captions directly into the video editing project.
- 4Manage timeline changesUse the standard undo command to revert any edits the AI makes to the timeline.
Written from the project's README. Every use case is checked against a quote from it.
How to install CartCut MCP
Formats verified 10 Nov 2026Install CartCut MCP in Claude Code
$ claude mcp add --env CARTCUT_MCP_TOKEN=<YOUR_CARTCUT_MCP_TOKEN> cartcut -- npx -y @cartesiancs/cartcut-mcpInstall CartCut MCP in Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"cartcut": {
"command": "npx",
"args": [
"-y",
"@cartesiancs/cartcut-mcp"
],
"env": {
"CARTCUT_MCP_TOKEN": "<YOUR_CARTCUT_MCP_TOKEN>"
}
}
}
}Install CartCut MCP in VS Code
.vscode/mcp.json
{
"servers": {
"cartcut": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@cartesiancs/cartcut-mcp"
],
"env": {
"CARTCUT_MCP_TOKEN": "<YOUR_CARTCUT_MCP_TOKEN>"
}
}
}
}Install CartCut MCP in Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"cartcut": {
"command": "npx",
"args": [
"-y",
"@cartesiancs/cartcut-mcp"
],
"env": {
"CARTCUT_MCP_TOKEN": "<YOUR_CARTCUT_MCP_TOKEN>"
}
}
}
}Install CartCut MCP in Codex
~/.codex/config.toml
[mcp_servers.cartcut] command = "npx" args = ["-y", "@cartesiancs/cartcut-mcp"] [mcp_servers.cartcut.env] CARTCUT_MCP_TOKEN = "<YOUR_CARTCUT_MCP_TOKEN>"
Install CartCut MCP in Gemini CLI
~/.gemini/settings.json
{
"mcpServers": {
"cartcut": {
"command": "npx",
"args": [
"-y",
"@cartesiancs/cartcut-mcp"
],
"env": {
"CARTCUT_MCP_TOKEN": "<YOUR_CARTCUT_MCP_TOKEN>"
}
}
}
}Configuration
CARTCUT_MCP_TOKENSecretOptional. CartCut's MCP token; read from CartCut's own settings when unset.
| Name | Description | |
|---|---|---|
| CARTCUT_MCP_TOKEN | Optional. CartCut's MCP token; read from CartCut's own settings when unset. | Secret |
Automated checks
Ran Nov 12, 2026Full report →Passed our automated checks. That is not a security audit: read the code and give the server the narrowest access it needs. How checks work
- Source repository linked and reachablePass
- Open source license (OSI)Pass
- Commit or release in the last yearPass
- Package name not close to a popular packagePass
- Remote endpoint uses HTTPSNot checked
- Destructive tools are labelledNot checked
- No hidden instructions to the modelPass
- No critical advisory in the package or its dependenciesPass
README
View on GitHub@cartesiancs/cartcut-mcp
Connects Claude, and any other MCP client, to the CartCut video editor running on your computer. Claude edits the project you have open: cutting from a transcript, adding captions, trimming and restaging clips, motion, transitions and effects. Every edit lands in CartCut's own undo history, so ⌘Z takes it back.
CartCut runs its MCP server inside the app, on http://127.0.0.1:9826/mcp.
This package is the stdio side of that connection: the client starts it, and it
relays every message to the app and back. It needs no configuration: it reads
CartCut's token from CartCut's own settings.
Requirements
- CartCut, open. Download it here.
- Node.js 18 or later, for
npx.
Install
Claude Code
claude mcp add cartcut -- npx -y @cartesiancs/cartcut-mcp
Add -s user to have it in every project. For the editing skill as well, which
is what makes Claude a good editor rather than one that merely has the tools,
install the plugin instead:
plugins/cartcut-editing.
If the skill is not installed, the bridge tells Claude where to read it.
Claude Desktop, Cursor and other clients
{
"mcpServers": {
"cartcut": {
"command": "npx",
"args": ["-y", "@cartesiancs/cartcut-mcp"]
}
}
}
Codex
codex mcp add cartcut -- npx -y @cartesiancs/cartcut-mcp
Configuration
Both variables are optional.
| Variable | Default | |
|---|---|---|
CARTCUT_MCP_TOKEN |
read from CartCut's settings | The token under the ⚡ icon at the bottom right of the CartCut window. |
CARTCUT_MCP_URL |
http://127.0.0.1:9826/mcp |
Where CartCut listens. |
Without CARTCUT_MCP_TOKEN, the token comes from CartCut's config.json:
- macOS:
~/Library/Application Support/cartcut-app/config.json - Windows:
%APPDATA%\cartcut-app\config.json - Linux:
~/.config/cartcut-app/config.json
Behaviour worth knowing
CartCut can restart underneath a session. Quit it, update it, open it again: the next call opens a new session with the app and carries on, and the client is told to re-read the tool list.
A call is never sent twice. If CartCut closes while a tool call is running, the call is answered with an error rather than retried, because it may already have changed the timeline. Look at the editor before asking again.
Nothing leaves the machine. The bridge only ever talks to the address above.
Troubleshooting
- "CartCut is not running, or its MCP bridge is off": open CartCut, then
reconnect (
/mcpin Claude Code). - "CartCut refused the token":
CARTCUT_MCP_TOKENis set to an old token. Unset it, or copy the current one from the ⚡ icon. - "There is no CartCut token": CartCut has never been opened on this account. Open it once.
npx -y @cartesiancs/cartcut-mcp --version prints the version. The bridge logs
to stderr and never to stdout, which carries the protocol.
License
MIT
CartCut MCP questions
How do I add CartCut MCP to Claude Code?
Run claude mcp add --env CARTCUT_MCP_TOKEN=<YOUR_CARTCUT_MCP_TOKEN> cartcut -- npx -y @cartesiancs/cartcut-mcp in your terminal. Add --scope user to make it available in every project.
How do I add CartCut MCP to Cursor?
Open ~/.cursor/mcp.json and add the server under mcpServers, using the Cursor config shown in the install section above.
How do I connect Claude to CartCut?
You can add it via Claude Code using 'claude mcp add cartcut -- npx -y @cartesiancs/cartcut-mcp' or by adding the server configuration to your Claude Desktop JSON settings.
Does my video data leave my computer?
No, the bridge only communicates with the local address where CartCut is running, ensuring nothing leaves your machine.
What happens if CartCut closes during a tool call?
The call will be answered with an error instead of being retried to prevent accidental duplicate edits to the timeline.
Where can I find my CartCut MCP token?
The token is located under the lightning bolt icon at the bottom right of the CartCut window.