An MCP video editor for CapCut is an MCP server that lets an AI app such as Claude or Cursor drive video editing work on your machine. The open source mcp-video-editor project is one example: it targets CapCut and OpenCut from Cursor, Claude, Codex, Antigravity, OpenCode, KiloCode, Qwen Code and ZCod, so you can run short-form edits from a prompt instead of clicking through a timeline.
raw clip
|
v
+---------------------+ +---------------------+
| AI app | MCP | mcp-video-editor |
| Claude / Cursor |----->| server |
+---------------------+ +----------+----------+
|
v
+---------------------+
| CapCut / OpenCut |
| project + render |
+----------+----------+
|
v
final short, ready to post
through Creator OS
What Is an MCP Video Editor for CapCut?
It is an MCP server that hands an AI agent editing actions inside a video project, instead of you clicking through a timeline.
MCP is the protocol Claude, Cursor and other AI apps use to reach outside tools. A video editing MCP exposes editing moves (cut, trim, add captions, place overlays, export) as tools the model can call. The AISEO-VirtualabDigital/mcp-video-editor repo is one implementation. Its README describes working with CapCut and with OpenCut as the editors it targets.
Two things make it worth a look:
- It is open source, so you can read the code before you run it.
- It is a repo you can audit. You can see exactly which commands the server runs before you trust it.
The project’s own framing is simple: it exists so you can drive CapCut and OpenCut edits from an AI app instead of doing every cut by hand.
How Does an MCP Video Editor Work With Claude and Cursor?
The AI app reads the server’s tool list, then calls those tools while you describe the edit you want.
Install the server, register it in your app’s config, restart, and the tools appear. From there the loop is conversational:
- You point the agent at a clip and describe the result: “make a 40 second vertical cut, hook in the first two seconds, captions on every spoken line.”
- The agent picks tools from the server and runs them.
- You review the project in CapCut or OpenCut and adjust by hand for taste.
Cursor is the better fit when you want to see commands and diffs as they happen. Claude Desktop is the better fit when you want to talk through the edit without looking at a terminal. Either way, the model is doing the sequencing, and the server is doing the file work.
MCP Video Editing Setup: Install the Server Step by Step
Setup is a config file entry plus a restart. Budget ten minutes the first time.
Clone the repo, install its dependencies, then register it as an MCP server in whichever app you use.
git clone https://github.com/AISEO-VirtualabDigital/mcp-video-editor
cd mcp-video-editor
# install per the repo's README, then register the server:
# Claude Code
claude mcp add video-editor <command from the README>
# Cursor, Windsurf, VS Code: mcpServers entry in the app's
# MCP settings file, pointing at the same command
Then check it loaded:
- In Cursor or Claude Code, ask the agent to list its tools and confirm the video tools are there.
- Run one harmless call first, such as reading clip metadata, before you let it render anything.
- Keep a copy of the source file until you have seen one clean export.
Follow the repo’s README for the exact command and flags, since they change between versions. For a longer look at what MCP connections can do once they are wired up, see our walkthrough of using Cursor with an MCP server.
Claude Video Editor GitHub: What the Repo Actually Gives You
You get an MCP server and its tool definitions, not a hosted editor and not a polished desktop app.
What that means in practice:
- You run it. The server runs on your machine, using your own hardware for the work.
- You read the code. Open source means you can audit what each tool does before you point it at client footage.
- You fix it. If a tool fails on your file format, the fix is a pull request away, not a support ticket.
The same open source, bring-your-own-machine pattern shows up in other Creator OS work. OPEN VIDEO EDIT is our open source set of Claude Code skills that turns a talking head clip into an animated short, rendering locally with Python and ffmpeg. Skills like fireship-style-short and split-animated-short handle captions, logos and b-roll cards. If your goal is a styled short rather than a CapCut project, that path may suit you better. Watch the demos on the KevBuildsApps YouTube channel and the launch video.
Claude Shortform Video Editor: Editing a Clip End to End
A shortform edit is a repeatable recipe, which is why an agent does it well. Here is one pass.
Prompt you can adapt:
Take clip_a.mp4. Target 9:16, 38 to 45 seconds.
1. Cut the first 1.5 s of silence.
2. Keep the strongest hook line in the first 2 s.
3. Add burned captions, one line at a time, max 6 words.
4. Add a progress bar at the bottom.
5. Export to out/clip_a_short.mp4, then show me a summary
of every cut you made and its timestamp.
The summary step matters. Ask for it every time. It gives you a cheap way to check the agent’s choices without scrubbing the whole timeline, and it teaches you which phrasing gets good results.
What the agent will not do is judge taste. Pacing, which take is funnier, whether the b-roll lands: those stay with you. Treat the server as a fast first assembly, then finish in CapCut by hand.
Video Editing MCP vs Manual Work: What Changes
The trade is your setup time and your machine’s cycles against the hours you would spend on every cut and caption by hand.
| Approach | What it needs from you | What you do yourself |
|---|---|---|
| Editing MCP server (open source) | A local install and the AI app you already use | Configure, debug, review each export |
| Clicking through a timeline | An editor and your time | Every cut, caption and export by hand |
| Both, in sequence | The MCP server plus an editor | Review the agent’s assembly, finish by hand |
Two honest tradeoffs. First, you are the support desk when a tool breaks. Second, the model calls cost tokens, so a long render session on a big model is not free. Run the cheap steps on a smaller model and keep the expensive model for the decisions.
Can You Use This MCP for Video Editing Alongside Creator OS?
Yes, and the split is clean: the editing MCP makes the file, and Creator OS ships it.
Creator OS is the publishing and engagement layer: 11 platforms plus Skool, one place to schedule, reply and read analytics, usable from the web app, the iOS app, the REST API, the CLI and a hosted MCP server for Claude, ChatGPT, Cursor and Codex. The two servers sit side by side in the same AI app.
One session can look like this:
- The editing MCP produces
out/clip_a_short.mp4. - You upload it once:
creatoros media:upload out/clip_a_short.mp4. - You post it:
creatoros posts:create --text "hook line here" --platforms instagram,tiktok,youtube --media <med_id> --scheduledAt 2026-03-04T18:00:00Z --timezone America/Toronto - You read the numbers back later and feed them into the next edit.
Step four is where it compounds. Instagram publishes Reel watch stats through Creator OS: reelsSkipRate, the share of viewers who swiped away in the first three seconds, plus average watch time. An agent can read those after posting and change the next cut. If skip rate is high, shorten the hook. Creator OS supplies the posting and the numbers; your prompt decides what to change. Setup for the hosted server is in the MCP docs.
FAQ
What is the best MCP video editor for CapCut?
The AISEO-VirtualabDigital/mcp-video-editor repo is the one built specifically around CapCut and OpenCut, and it is open source so you can read it before you run it. If your goal is animated shorts rather than CapCut projects, Creator OS also publishes OPEN VIDEO EDIT, an open source set of Claude Code skills that render locally with Python and ffmpeg. Neither one comes with a support contract, so read the code before you point either at client footage.
How do I connect a video editing MCP to Claude or Cursor?
Install the server locally, then add an entry for it in your app’s MCP settings: Claude Code takes a claude mcp add command, while Cursor, Windsurf and VS Code take an mcpServers entry with a type, a command and optional env. Restart the app and confirm the video tools appear in the tool list before you run anything that renders. The exact command lives in the repo’s README, since it changes between versions.
Can Claude edit CapCut projects directly with an MCP server?
It can drive a CapCut workflow through the server’s tools, which is what the mcp-video-editor project is built for. Claude is not opening the CapCut interface and clicking; it is calling editing tools that act on your project and media files. What it can do depends entirely on which tools the server exposes, so list them first.
Is the mcp-video-editor GitHub repo safe to run locally?
Running any MCP server locally means granting a process access to your files and the commands it needs, so treat it like any script you download. Clone it, read the tool definitions and any shell commands it runs, test on a throwaway clip in a folder that holds nothing sensitive, and keep backups of source media. Open source makes that audit possible, and doing the audit is still your job.
What else can a Claude shortform video editor automate?
Beyond the cut itself, the same session can write the caption and check it against each platform’s limits with creatoros validate:post-length --text "...", upload the finished file once and reuse the media id, and schedule across Instagram, TikTok and YouTube Shorts in one call. After publishing, an agent can pull per-post analytics and use Reel watch stats to decide what to change in the next edit.
Get started
Build the edit with a local MCP server, then hand the file to Creator OS to publish and measure. A free trial starts at creatoros.ca/sign-up, and the Creator plan is $14.99/month for 11 platforms plus Skool, with the API, the MCP server and the CLI included. Start with the social media MCP server overview, then watch the walkthroughs on the KevBuildsApps YouTube channel.
Get started for $0
Free trial for new subscribers. Cancel anytime.