Most video tools now say they work with Claude. But what exactly does ‘working with Claude’ mean? With one tool, Claude can open your project, move a single callout, and update the help article that goes with it. With another, Claude can send a prompt, wait for an MP4, and do nothing more.

The difference comes from the MCP server, the connector a vendor builds so an AI assistant can use its product. The vendor picks which features go into it. Some put in most of what their app does. Others put in two or three actions.

In July, a team lead at a sales enablement software company told us their team already ran parts of its design work through Claude. Their first question about video wasn't about quality or price. They wanted to know what the agent could actually do inside our product, and whether it could work alongside the other MCP servers they used.

This guide answers that question for eight video MCP servers. We read each vendor's documentation, list of tools, and pricing page. We didn't test every tool by hand, and we say so where it matters.

What is a video editing MCP server?

A video editing MCP server is a connector that lets an AI assistant, such as Claude, ChatGPT, or Cursor, use a video tool on your behalf. The assistant can create a project, add text or a voiceover, change a clip, and export a video while you describe what you want in plain language.

MCP stands for Model Context Protocol, an open standard Anthropic published in November 2024. A vendor's MCP server publishes a list of actions, which MCP calls "tools." Each tool has a name, such as split_clip, and a short description. Your assistant reads the list and decides which tools to call to get the job done.

The standard doesn't say which actions a video tool should offer. That's why two "MCP-compatible" video tools can give your agent very different amounts of control. Our explainer on what a video MCP server does and what to check before you connect one covers the basics in more depth.

How is using a tool through MCP different from using the app?

When you use a video app yourself, you can see the screen and reach every menu. When an agent uses the same product through its MCP server, five things change.

The agent can only take the actions on the list. If the vendor didn't build a tool for something, the agent can't do it, even when the app can. Synthesia's app edits videos, but its MCP server can only create and generate them, so it can't change a video that already exists. Descript's MCP server can't give you a video file until you publish the project as a link.

The agent can't see the screen. In the app, you'd notice a label covering a button. The agent only knows what the tools send back, which is usually positions and timings as numbers. Unless a tool returns a picture of the frame, the agent places things without seeing them.

Your plan still applies. Actions taken through MCP use the same credits, render minutes, and plan limits as clicks in the app. Features locked to a higher plan stay locked. Supademo's MCP server needs its Scale plan, and Clueso's needs a paid plan to export video.

The result ends up somewhere. Some servers save the agent's work into a project you can open in the app and keep editing by hand. Others hand back a finished file, so changing it means asking the agent to make it again.

Someone has to approve the connection. Hosted servers ask you to sign in with your account and then act inside your workspace. Local servers run on a computer and may be able to reach its files. One customer education lead we spoke to needed their security team's approval before anyone could connect.

Three ways a video MCP server can work

Start with what you want to have when the agent is done.

If you want a project your team can open and keep changing, choose a server that edits a project in a video tool. Your team can then change it by hand or by asking the agent again. If you want the video stored as code that a developer reviews, choose a code-to-video framework. If you want a short clip from a sentence and don't mind generating it again to change anything, choose a generator.

What to look for in MCP servers

Can the agent change one part without redoing the whole video? Read the vendor's list of tools before the marketing page. Tool names tell you a lot. A server built around generating video has names like create_video and get_status. A server built around editing has names like get_project, update_elements, and split_clip. If nothing on the list changes something that already exists, the agent can make a first draft but can't revise it.

Can the agent see what it made? Some servers show a preview, but it's a video player for you to watch, not something the agent can look at. Ask whether any tool sends an image of a frame back to the agent. That's what lets it notice a label covering a button and fix it before you export.

Does the voiceover stay in sync with the screen? In a product video, the voiceover says "click Save" at the moment the cursor reaches Save. If the voice and the picture aren't linked, every script change shifts the timing and someone has to fix it by hand. Generators avoid this by making the whole clip again, which also changes the parts you liked.

What's left for you to do in the app? Ask the vendor which steps still need a person: recording the screen, final polish, downloading the file, approving changes. A short list means the agent is doing real work. A long one means you're mostly typing instructions into a chat window.

Editorial methodology

Inclusion criteria. A server made the ranked list only if it met all four of these:

• The vendor or project documents an MCP server or an official agent integration.
• It works with Claude, Codex, or any other third-party LLM.
• It creates or modifies a video file.
• It shipped a release or code change in the past six months.

What we evaluated. Each vendor's MCP documentation, list of tools, plan limits, and pricing page, read between October 12 and October 14, 2026, plus each open-source project's README and license. Nobody tested every tool by hand for this guide. Where a vendor doesn't publish something, such as a tool count or a price per render, we say so rather than guess.

Servers that edit a project

With these, the agent works on a project you can open and keep editing. Choose one of them if you'll still be changing the video next month.

1. Clueso — best for teams making product, training, and launch videos from Claude

What it is. Clueso is the AI video and documentation platform. Teams use it to turn screen recordings, documents, and slides into product and training videos, each with a step-by-step guide. The Clueso MCP server connects your assistant to your Clueso workspace, so the agent works on the same projects your team opens in the editor. It works with Claude, Claude Code, ChatGPT, Codex, Cursor, Gemini CLI, VS Code, and Antigravity.

What your agent can do through the MCP. The server has more than 40 tools. With them, the agent can:

• Start a project from a screen recording, a slide deck, or a blank canvas, and add, split, or reorder clips.
• Change one element without touching the rest. That covers text, callouts, arrows, zooms, spotlights, and blurs that hide customer data, among ten element types.
• Look at a picture of any clip with get_clip before exporting, so it can check that a callout sits where it should.
• Write an AI voiceover and link words in it to moments in the recording. auto_sync finds the moments where the screen changes, and add_sync_point pins a word to one of them. If you rewrite the script, the zoom still lands on the right button.
• Write and illustrate the step-by-step guide that goes with the video, using screenshots from the same project.
• Apply a Clueprint, Clueso's saved template for brand, layout, and editing style, so the agent doesn't need your colors and fonts restated in every prompt.
• Follow any of 90 free video skills, ready-made instructions for jobs like crm-training-video or batch-multi-language-export.
• Generate motion graphics and images inside a clip, translate the voiceover, captions, and on-screen text, and export the finished video.

What you still do in the app. Recording your product. The MCP server works with recordings a person makes, or you can use Clueso Agents, which records the product for you. Final touches are optional, and they happen in the same project in the Clueso editor. Teams delivering through an LMS can also use SCORM export on the Enterprise plan.

Pricing. The MCP server is included on every plan, including Free. Exporting video and advanced branding need a paid plan. Solo is $40/month and Business is $120/month, both billed yearly. Enterprise pricing is custom.

Why teams use it. The Teamworks customer education team supports more than 20 products with three people. Most of their videos now start in Claude. They name the Clueprint, upload the raw recording, and describe the edits, callouts, and format, then open Clueso only for final touches. A video used to take four to five hours across four tools. It now takes 20 to 30 minutes, and the team went from three to five videos a month to 10 to 15. Read the Teamworks story.

Where it falls short. The MCP server can't record your product by itself, so a person records or you add Clueso Agents. It also isn't built for camera footage, interviews, or color grading. For those, Descript or Resolve is a better fit.

Bottom line: the strongest choice when an agent needs to keep editing your product or training videos after the first version.

Compare: Clueso MCP · What to check before you connect a video MCP server

2. Descript — best for editing interviews and talking-head video by editing the transcript

What it is. Descript lets you edit audio and video by editing the transcript. Delete a sentence from the text, and the matching footage is cut.

What your agent can do through the MCP. Descript's official server lets the agent:

• Import audio and video into Descript.
• Ask Underlord, Descript's built-in AI editor, to make edits described in plain language. Descript labels Underlord as beta.
• Organize projects and folders, and publish a project as a shareable web link.
• Export the transcript as text, Markdown, HTML, RTF, or SRT captions.

What you still do in the app. Precise edits on the timeline. Downloading the video file, since the MCP server can only publish a link first. Descript's AI image and video generation, which isn't included in the Claude connector.

Pricing. Included for all paying Descript users at no extra cost. It uses your plan's media minutes and AI credits. Hobbyist starts at $16/month billed yearly, or $24 billed monthly.

Why teams use it. For podcasts, interviews, and talking-head videos, editing a transcript is faster than working on a timeline, and the MCP server brings that into your assistant.

Where it falls short. Descript doesn't publish its list of MCP tools, so you can't check exactly what the agent can change before you connect. In Claude, it needs code execution and network access turned on. It doesn't write step-by-step guides, and it doesn't link narration to moments on screen.

Bottom line: the best server here for footage of people talking. For product walkthroughs, choose a different tool.

Compare: Clueso vs Descript · Descript alternatives

3. Shotstack — best for making video inside your own app

What it is. Shotstack is a video editing API for developers. You describe a video as a timeline written in JSON, a structured text format, and Shotstack's servers turn it into an MP4.

What your agent can do through the MCP. Shotstack's hosted server has 15 tools. The agent can:

• Build a timeline and render it into a video.
• Save templates with fill-in fields, then render many versions of them.
• Generate AI clips and voice, which are billed separately.
• Open a preview in the chat with the studio tool.

What you still do in the app. Review the preview and click Render yourself. The preview is for you, not the agent. Everything else the MCP server does is also available through Shotstack's API, so developers can build the same steps into their own product.

Pricing. Pay as you go is $10 for 50 minutes of rendering. Plans run from Starter at $39/month for 250 minutes to Volume at $499/month for 10,000 minutes. A free test environment renders with a watermark, up to 10 minutes. AI generation costs extra.

Why teams use it. Developers who already create video from their own app get an agent that writes the same timelines they'd write by hand. A person still approves each video before it spends render credits.

Where it falls short. The agent can't see the preview, so it places things by numbers alone. There's no screen recording, no link between narration and the screen, and no written guide. Voiceover comes from paid AI generation rather than a built-in voice track.

Bottom line: the most complete hosted timeline API with an MCP server. Choose it when video is a feature of your own product.

4. Creatomate — best for many versions of one template

What it is. Creatomate makes videos from templates. You design a layout once, then swap the text, images, and clips to produce many versions.

What your agent can do through the MCP. Creatomate's own hosted server has seven tools. The agent can create and update templates, render videos from them, and check on renders.

What you still do in the app. Most teams design templates in Creatomate's visual editor. The agent then fills them in and renders them.

Pricing. Renders use your plan's credits.

Why teams use it. Marketing and social teams making hundreds of personalized versions of one design.

Where it falls short. With seven tools, the agent can change the template but not much else. There's no voiceover track, no screen recording, and no written guide.

Bottom line: a small, dependable server for template versions. Too narrow for editing a walkthrough.

5. DaVinci Resolve Studio — best for finishing footage in a professional editor

What it is. Blackmagic Design's professional app for editing, color, visual effects, and audio. Version 21.1, announced September 8, 2026, includes a built-in MCP server. You turn it on from File > Setup AI Assistants.

What your agent can do through the MCP. Blackmagic says Claude, Claude Code, or Codex can analyze a project, organize media, cut highlight edits, and batch render. The server works through Resolve's scripting interface, so the agent can reach what scripting can reach.

What you still do in the app. Creative work like color grading and fine-tuning cuts, plus anything the scripting interface doesn't cover. You'll also want to review what the agent changed. Blackmagic advises trying it on copies of projects first.

Pricing. The MCP server needs the Studio edition, a one-time $295 license. The free edition doesn't include it. (Blackmagic and press coverage, September 2026)

Why teams use it. Editors who already finish their work in Resolve can hand repetitive jobs, like logging footage, picking selects, and queuing renders, to an assistant.

Where it falls short. Blackmagic doesn't publish how many tools the server has or what stops it from making destructive changes. Resolve is a desktop app with a steep learning curve, so it won't help a team without an editor. There's no AI voiceover or written guide.

Bottom line: the right server if a human editor finishes the work in Resolve anyway. More than most teams need.

6. Kinocut — best for free edits to video files on your own computer

What it is. A free, open-source MCP server that runs on your computer. It uses FFmpeg, the command-line program behind most video conversion, to edit files.

What your agent can do through the MCP. Version 1.16.1 has 203 tools. The agent can trim, join, convert, and caption videos, and level the audio. Kinocut checks each request before it runs, so a bad command stops instead of overwriting a file. It asks for your approval before anything is published, and it keeps a record of every change made to a file.

What you still do in the app. There is no app. The results are files on your computer, which you review in a video player.

Pricing. Free. It needs Python 3.11 or later and FFmpeg installed. (Kinocut GitHub repository, October 2026)

Why teams use it. Developers who want an agent to trim, join, and caption files without a subscription or uploading anything.

Where it falls short. It works on one computer, with no shared project for teammates. There's no AI voiceover. And 203 tools is a lot for an agent to choose from, which can mean more wrong choices when a request is vague.

License. Apache-2.0

Bottom line: the safest of the FFmpeg-based servers. It edits files you already have, so don't expect it to make a video from nothing.

Other servers you'll find in directories

Supademo has an official MCP server with more than 80 actions for interactive click-through demos. It asks for your approval before making changes and needs the Scale plan ($50/month) or above. It makes interactive demos rather than video files.

Premiere Pro has community-built MCP servers, with anywhere from 74 to more than 1,000 tools, that connect through a plugin on your computer. Adobe doesn't offer an official one.

CapCut has no official server either, but the open-source VectCutAPI project can create CapCut projects through MCP.

Jumper adds footage search to Premiere, Resolve, Final Cut Pro, and Avid. OpusClip and Reap cut long videos into short captioned clips.

Canva's MCP server can export animated designs as MP4.

Comparison table

Which should you choose?

Your team makes product walkthroughs, training, or launch videos and updates them with every release: Clueso.

You need a step-by-step guide alongside each video: Clueso, the only server here that writes one.

Your developers want videos stored as HTML: HyperFrames.

Your developers already write React, or you make data-driven video at scale: Remotion.

Video is a feature inside your own app: Shotstack, or Creatomate if it's mostly versions of one template.

You edit interviews, podcasts, or talking-head footage: Descript.

A professional editor finishes everything in a desktop editor: DaVinci Resolve Studio.

You want free edits to files on your own computer, with an approval step: Kinocut.

You need an AI presenter on screen: HeyGen or Synthesia.

Is Clueso right for you?

Pick Clueso if your videos show software or a process, more than one person makes them, and they need updating whenever the product changes. It also fits if your team already works in Claude or ChatGPT and wants the video and its step-by-step guide to come out of the same conversation. The Teamworks setup, with one Clueprint per product and most videos started in Claude, is a good model to copy.

Pick something else if you're editing camera footage (Descript or Resolve), building video into your own product (Shotstack, Remotion, or HyperFrames), or need an AI presenter (HeyGen or Synthesia). If you want an agent to record your product as well as edit the video, look at Clueso Agents rather than the MCP server.

Related reading: what to check before you connect a video MCP server · setting up the Clueso MCP · 90 video skills for AI agents

FAQ

Is there a video editor with MCP support?

Yes, several. Clueso, Descript, Shotstack, Creatomate, and DaVinci Resolve Studio all have official MCP servers that edit a project rather than only generate a clip. Resolve's is built into version 21.1 and needs the paid Studio edition. Adobe Premiere Pro and CapCut have community-built servers but no official one.

What is the best video MCP server for AI agents?

It depends on what you want the agent to leave behind. For product and training videos your team keeps editing, Clueso is the strongest fit. The agent edits the same project people open by hand, and it can look at the result before exporting. For video as code, choose HyperFrames. For video made inside your own app, choose Shotstack.

Which MCP servers turn screen recordings into videos?

Clueso is built for this. You upload a recording, and the agent adds an AI voiceover, zooms, and callouts, keeps the voiceover in sync with what happens on screen, and writes the matching step-by-step guide. Kinocut and other FFmpeg-based servers can trim and caption a recording, but they don't add narration or structure.

How does Clueso's MCP compare to HyperFrames for letting an AI agent create videos?

With HyperFrames, the agent writes HTML that becomes a video, so the result is code a developer maintains. With Clueso, the agent builds and edits a project in a video editor, with AI voiceover, narration synced to screen recordings, brand templates, and a step-by-step guide. Choose HyperFrames for motion graphics your developers own. Choose Clueso for product and training videos that people outside engineering also edit.

Does CapCut have an MCP server?

Not an official one. The open-source VectCutAPI project can create CapCut and JianYing project files and includes an MCP server. Several community servers are built on top of it, and you run them on your own computer.

Can AI edit my videos for me?

Yes, if your video lives in a tool whose MCP server can change an existing project. Ask the agent to shorten a section, move a callout, or swap the voice, and it changes only that part. A generator makes the whole clip again instead.