RendGet started

Blog · November 17, 2026

MCP Video Editing: How to Edit Video With Any AI Agent (2026)

MCP video editing means an AI agent (Claude, ChatGPT, Codex, Cursor and others) edits video by calling tools on a video editor's MCP server. The agent never touches the file itself. It asks the server to transcribe, cut, caption, render and export, then reads back the results. Pick a server by what you want returned (an editable project, a rendered MP4, or generated clips), and connect it once.

This guide explains the categories of servers, gives you a decision table, and shows exact setup for Rend's server at https://rend.ai/api/mcp in every major client.

How an MCP video edit actually works

The Model Context Protocol is a standard way for an AI app to discover and call another app's tools. A video server publishes tools like "upload media", "transcribe", "edit the composition", "render". Your agent chooses which to call based on your request.

A typical edit loop looks like this:

  1. You describe the video.
  2. The agent calls tools to read your media (transcript, frames, metadata).
  3. It changes a composition or project on the server.
  4. It starts a render, which runs as a background job.
  5. It checks the job until the file is ready, then gives you the link.

Steps 3 and 4 are where servers differ most. Some edit a timeline you can open later. Some only run FFmpeg commands on your disk. Some only generate new clips and can't touch your footage.

The four kinds of video MCP servers

KindWhat you get backExamples (Nov 2026)Pick it when
Hosted editorAn editable project plus exportsRend, Descript, Kapwing, invideo, OpusClip, TellaYou want to review and tweak the agent's work
Template rendererMP4 from JSON or a templateShotstack, CreatomateYou build automated, data-driven videos
Generation-onlyNew clips, images or audioVarious generation and voice toolsYou need assets, not edits
Local toolsProcessed files on your machineFFmpeg MCP servers, desktop-editor bridgesYou want free, local, scriptable file work

One note on Remotion, because it shows up in a lot of older articles: Remotion's hosted MCP server was shut down on October 26, 2026, per Remotion's own docs. Remotion now points agents to its Agent Skills instead. If a tutorial tells you to add the Remotion MCP, it's stale. More on that in Claude Code video editing.

What makes Rend's server different

Most hosted servers expose an editor. Rend exposes a whole video agent: a crew of specialists that takes an idea through script, voice, music, generated shots, captions and export, with every result on a real timeline. When you connect an agent to Rend, it gets storage, rendering and every tool, including:

  • Cuts on exact transcript words, transcripts with speakers.
  • Word-timed captions, always editable, in your font and look.
  • Titles, motion graphics, Lottie and 3D graphics.
  • Narration in a voice you approve, original music, sound effects and a ducked mix, each on separate tracks.
  • Generated shots, images and b-roll; web research.
  • Brand kits reused every episode, long videos built in chapters in parallel.
  • A separate Linux computer per project with Blender and LibreOffice for the odd jobs.
  • Review links where people draw on the frame and the agent fixes each note.

The tools page lists it all. Editing is free; you pay when Rend renders or generates something (pricing).

Setup for every client

Rend's server uses Streamable HTTP and OAuth 2.1 with dynamic client registration and PKCE. Most clients only need the URL, then open a Rend page in your browser to sign in and approve. The client stores and refreshes its own credentials. Rend is listed in the official MCP Registry as ai.rend/rend.

Claude Code

claude mcp add --transport http rend https://rend.ai/api/mcp

Run /mcp, choose rend, authenticate. Use the latest Claude Code so tools load on demand.

Codex

codex mcp add rend --url https://rend.ai/api/mcp
codex mcp login rend

Claude (web and desktop)

Add a custom connector in Claude's settings with https://rend.ai/api/mcp, then sign in when prompted. Full walkthrough: Claude video editing.

ChatGPT

On the web, open Plugins, click plus, choose Add custom MCP server, enter the URL and pick OAuth (OpenAI's current flow; Developer mode guides are out of date). Details in ChatGPT video editing.

Cursor and other clients

Add a custom remote MCP server with the same URL. In Cursor that's an entry in your MCP config:

{
  "mcpServers": {
    "rend": { "url": "https://rend.ai/api/mcp" }
  }
}

Clients without OAuth

Create a personal token in Studio's agent connection settings and store it in the client's credential settings for bearer auth. Never paste it into a prompt or a shared project. For stdio-only clients there's a bridge script; the connection guide has the exact commands.

A first session, step by step

  1. Make or pick a project. Tell the agent the brief, then confirm aspect ratio and frame rate before it starts.
  2. Get media in: upload it, or give direct HTTPS links to files. Chat attachments only work if your client passes them to tools.
  3. Ask for a checkpoint: "outline first" or "show me the script".
  4. Approve, and let it build. Generation and processing return background jobs, which external clients follow with rend_get_job.
  5. Open the Studio link it returns. Same clips, layers, audio and captions. Fix anything by hand.
  6. Export again to bake changes into a new MP4. Earlier sources and exports remain.

Prompts worth stealing

  • "Make a 45-second vertical recap of this 20-minute talk. Keep the line about pricing. Captions bottom third, no music for the first 3 seconds."
  • "Build a 10-minute video essay on the history of the shipping container. Research first, outline, then a script I approve. Use our brand kit." (See the video essay script template.)
  • "From this product page, make three hooks, then ads in 9:16, 1:1 and 4:5." (See video ad sizes.)
  • "Apply every note on the review link. List any you skipped."

Choosing: a quick decision table

You wantUse
A finished, original video from an idea, with voice and musicRend
Quick transcript trims inside an existing Descript libraryDescript
Short clips pulled from a long recordingOpusClip or Rend
Batch renders from spreadsheet dataA template renderer
Free local conversions and trimsAn FFmpeg MCP server
Code-built motion graphicsRemotion with its agent skills

Nobody needs all of these. Most people need one hosted editor plus, occasionally, a local tool. If you're comparing agents more broadly, the 2026 roundup of AI video editing agents and what a video editing agent is go deeper.

Risks to know about

MCP servers can take real actions, and OpenAI and Anthropic both warn to connect only servers you trust. Specific to video: renders and generation cost money on most services, so ask your agent to show plans before it renders. And local servers that run shell commands on your computer deserve the same scrutiny as any script you'd download.

FAQ

What is a video editing MCP server?

A service that exposes video tools (upload, transcribe, cut, caption, render) through the Model Context Protocol, so an AI agent can call them on your behalf.

Which AI agents support MCP video editing?

Claude, Claude Code, ChatGPT, Codex and Cursor all support remote MCP servers. Exact steps differ per client; see the setup section above.

Is there an official Remotion MCP server?

Not anymore. Remotion shut down its hosted MCP on October 26, 2026 and recommends its Agent Skills.

Do I need FFmpeg or a GPU for Rend's server?

No. Rend's server is hosted. Rendering happens on Rend, and the stdio bridge doesn't require local FFmpeg either.

Can I edit the agent's work myself?

With Rend, yes. Every project opens in Studio with the same timeline the agent built.

Where is Rend in the MCP Registry?

It's listed as ai.rend/rend in the official MCP Registry.

Ready to try one? Create a free Rend account, connect your agent with the command for your client, and ask for a short cut of footage you already have.

Give Rend the job.

Editing is free. You pay only when Rend renders or generates something.