Skip to content

Guide

MCP video editing means an AI assistant controls an editing tool through a connector

Updated

MCP video editing is when an AI assistant such as Claude or ChatGPT drives a video tool through a Model Context Protocol connection. Here's how it works, who it's for and when you can skip it.

In short

  • MCP video editing means an AI assistant, such as Claude or ChatGPT, controls a video editing tool through an MCP connection.
  • MCP, the Model Context Protocol, is an open standard for connecting AI apps to outside tools. Its own docs compare it to a USB-C port.
  • A video MCP server is either local, running ffmpeg on your computer, or remote, run by an editor such as Descript, OpusClip or Kapwing.
  • MCP is worth setting up if you already work in Claude, ChatGPT or a coding tool and want your video tool inside that chat.
  • For an AI edit with no setup, an editor with its own agent, such as Cutroom's Magic Editor, makes the edit itself.

What is MCP video editing?

MCP stands for Model Context Protocol. Its official docs call it an open-source standard for connecting AI apps to outside systems, like a USB-C port for AI.

MCP video editing is that standard used for video. You ask your assistant for an edit, and it calls a video tool through MCP to make it.

How MCP video editing works

The parts named in the Model Context Protocol's own docs, applied to a video edit.

  1. You ask your assistant

    In Claude, ChatGPT or a coding tool

  2. The assistant picks a tool

    From the tools the MCP server lists

  3. The MCP server runs it

    On your computer or at the editor's company

  4. The editing tool changes the video

    ffmpeg, or an editor such as Descript

  5. The result comes back

    A file, a link or a project to finish

What an MCP server for video does

An MCP server is a small program that lists the actions it offers, such as trimming or adding captions. The assistant picks one, and the server carries it out.

Video MCP servers come in two kinds. Local ones run on your computer, usually around ffmpeg. Remote ones are run by an editing company and edit in its cloud.

  • A local server keeps the video on your computer, but you install Python or Node and ffmpeg yourself. Kinocut needs Python 3.11 or newer.
  • A remote server needs a sign-in to the editor, and its AI edits use that plan's credits. Descript says the connection itself costs nothing extra.
  • OpenAI says it doesn't develop or verify custom MCP servers you add to ChatGPT, so only connect ones you trust.

Four video MCP servers, November 2026

Each from its maker's own page, checked 12 November 2026.

ServerRunsWhat it doesWhat you get back
Kinocut (was mcp-video)On your computerTrims, joins, adds text and audio with ffmpegA new video file
DescriptDescript's cloudAny edit Descript's AI can makeA Descript project, not an MP4
OpusClipOpusClip's cloudClips long videos, captions and reframes themShort clips
KapwingKapwing's cloudSubtitles and b-rollA project in your Kapwing workspace
  • Server: Kinocut (was mcp-video)

    Runs
    On your computer
    What it does
    Trims, joins, adds text and audio with ffmpeg
    What you get back
    A new video file
  • Server: Descript

    Runs
    Descript's cloud
    What it does
    Any edit Descript's AI can make
    What you get back
    A Descript project, not an MP4
  • Server: OpusClip

    Runs
    OpusClip's cloud
    What it does
    Clips long videos, captions and reframes them
    What you get back
    Short clips
  • Server: Kapwing

    Runs
    Kapwing's cloud
    What it does
    Subtitles and b-roll
    What you get back
    A project in your Kapwing workspace

How Cutroom's Magic Editor edits a video like this

Jacob is the AI editor inside Cutroom's Magic Editor: it edits your video for you, then changes it when you ask in a chat.

BEFORE · YOUR TAKESoum,I open thedashboard… (pause)Make my videoJacob edits itAFTER · JACOB'S FIRST EDITdashboard.pngDASHBOARDSo I open the dashboard.cut outyour fileon the word
A drawing of an example clip, not a customer video. What it shows is what the Magic Editor does: cuts the ums and pauses, puts your own screenshot on the word you say, sets key words big and adds music.

The 4 styles, as example edits

Steps

Three steps, four angles

Skit

Hook, hard stop, demo card

Verdict

The tester's verdict

Poster

Typography does the talking

Hover or tap an example to play it, without sound. These are example edits of each style; the presenters in them are generated with AI.

  1. Your take: one video of you talking, up to 3 minutes
  2. Your style: one of 4: Steps, Skit, Verdict or Poster
  3. Your files: screenshots, a screen recording or a logo, if you have them
  4. Your goal: what the video is for, in a few words, if you like
  5. Jacob makes it: normally 10 to 15 minutes, and you get an email
  6. Your edit: ask Jacob for changes in a chat, then export

Then you ask Jacob, for example

  • “Make this about 10 seconds shorter”
  • “Put a clip on the strongest sentence”
  • “Change the music to something calmer”

Each change is a new version, and going back is free. First cut 250 credits, each change 40, questions free, export 30 credits per started minute.

How to connect a video MCP server

Most people use a ready-made connector from the assistant's own directory. Adding your own server takes a web address or a local install.

  • In Claude, Anthropic says you open Customize, then Connectors, press + and connect a listed one such as Descript.
  • For your own server in Claude, you choose Add, then Custom, then Web, and enter the server's address. Free plans get one custom connector.
  • Anthropic says connectors work across Claude, Claude Desktop, Claude Code and its API. Local servers run in apps such as Claude Desktop and Claude Code.
  • In ChatGPT, OpenAI's docs say you open Plugins, choose Add custom MCP server and enter its address and sign-in details.
  • OpusClip says its server works in Claude, Cursor and ChatGPT, and you sign in once with your OpusClip account.

Who needs MCP for video editing

MCP helps when the assistant is already where you work. It adds little if you only want one video edited.

  • You need it if you spend your day in Claude, ChatGPT, Cursor or Claude Code and want the video tool in the same chat.
  • You need it if you already edit in Descript, OpusClip or Kapwing and want to direct that editor in plain words.
  • You need it if you're a developer building something repeatable, such as cutting the same kind of clip every day.
  • You don't need it if you film a clip of yourself talking and just want it edited. An editor with its own agent does that directly.

An MCP setup and an editor with its own agent

Two ways to get an AI edit. The MCP route needs an assistant, a server and an editing tool.

  • What you set up

    Assistant plus MCP server
    An assistant, a connector or a local server, the editing tool
    Editor with its own agent
    Nothing; you upload the clip
  • Where you ask

    Assistant plus MCP server
    In Claude, ChatGPT or a coding tool
    Editor with its own agent
    In the editor's own chat
  • Who does the edit

    Assistant plus MCP server
    The tool the server controls
    Editor with its own agent
    The editor's own AI
  • Best for

    Assistant plus MCP server
    People already working in an assistant
    Editor with its own agent
    A clip of you talking, edited start to finish
Assistant plus MCP serverEditor with its own agent
What you set upAn assistant, a connector or a local server, the editing toolNothing; you upload the clip
Where you askIn Claude, ChatGPT or a coding toolIn the editor's own chat
Who does the editThe tool the server controlsThe editor's own AI
Best forPeople already working in an assistantA clip of you talking, edited start to finish

Getting an AI edit without MCP

An editor with its own agent has the AI built in, so there's no assistant or server to connect.

Cutroom's Magic Editor turns one clip of you talking into a finished vertical video. Jacob, its AI editor, makes the whole edit, and you change it by asking in a chat.

  • You upload the clip and your logo and press "Make my video". It normally takes 10 to 15 minutes, and you get an email when your first cut is ready.
  • Then you type changes in plain words, such as "Keep my pauses, it feels too rushed". A message that changes the video costs 40 credits; a question is free, up to 20 a day.
  • It takes one clip up to 3 minutes and returns one vertical MP4 at 1080 by 1920. For long videos or podcasts, a remote MCP editor such as Descript fits better.
  • Cutroom is not affiliated with OpenAI, Anthropic or Google.

A physiotherapist's one-point clip, before and after

No connector, no server: drawn as Jacob's first edit in the Poster style.

Drawing of an example clip, not a customer video

Before · the clip as filmed

A 30-second phone video of a physiotherapist making one point about sitting at a desk

“Um, okay. Your back doesn't hurt because you sit badly. It hurts because you sit still. … (pause) Get up every thirty minutes, like, even, uh, even for one minute.”

struck through = what Jacob cuts

After · Jacob's first edit, Poster style

yourSTILLBACK1

“Your back doesn't hurt because you sit badly. It hurts because you sit still. Get up every thirty minutes, even for one minute.”

  • On “still”: the words sit still stacked big in three fonts
  • On “badly”: a black-and-white moment on the word
  • On “thirty”: the number 30 in neon
  • Under the whole video: your logo closing the video

Then you ask Jacob

  1. “Keep my pauses, it feels too rushed”v2the same edit with your pause before the last line kept
  2. “Change the music to something calmer”v3calmer music under the same words

Questions people ask

What does MCP stand for?
Model Context Protocol. It's an open-source standard for connecting AI apps such as Claude and ChatGPT to outside tools and data.
Can Claude edit videos with MCP?
Yes, through a connector to a video tool. Claude asks the tool, such as Descript or a local ffmpeg server, to make the edit. Claude doesn't render the video itself.
Does ChatGPT support MCP for video editing?
Yes. OpenAI's docs say you can add a custom MCP server under Plugins, and OpusClip and Descript say their servers work with ChatGPT. OpenAI doesn't verify custom servers.
Is MCP video editing free?
The protocol is free and open source, but the parts around it may cost money. Your assistant may need a paid plan, and remote editors use their own plan's credits for AI edits.
What is the difference between a local and a remote MCP server?
A local server runs on your computer, so the video stays there. A remote server runs at the editor's company, and you sign in to use it.
Do I need MCP to use Cutroom?
No. You upload your clip to Cutroom and Jacob, its own AI editor, makes the edit. There's nothing to connect. Cutroom is not affiliated with OpenAI, Anthropic or Google.

Skip the connectors: upload one clip and Jacob makes the whole edit. Your first video is free.

Try for freeYour first video is free, no card, and you have 7 days to make it. It carries a Cutroom mark; Lite at $19.99/month removes it