17:02Andy Diep · Tutorial
ChatGPT + DaVinci Resolve: The Full AI Editing Setup
A video editor wires ChatGPT's Codex into DaVinci Resolve through an MCP server, then lets it build a full first-pass cut on its own, mistakes included.
August 7thThe argument in one line.
Connecting free DaVinci Resolve to ChatGPT through an MCP server lets a $20-a-month ChatGPT plan handle the transcription and captioning that Resolve normally locks behind its $300 Studio license, for well under a dollar per edit.
Read if. Skip if.
READ IF YOU ARE…
- You edit in DaVinci Resolve's free version and want automatic transcription and captions without paying for Studio.
- You already use ChatGPT Plus or Claude and want it to control a desktop editing app, not just write scripts or copy.
- You're comfortable installing a GitHub repo and running a short setup conversation to connect an AI agent to local software.
SKIP IF…
- You already own DaVinci Resolve Studio, where captioning and transcription work natively with no workaround needed.
- You want a zero-setup, one-click tool rather than configuring an MCP server and reviewing a permission plan yourself.
The full version, fast.
DaVinci Resolve's free version can't transcribe or caption footage, that's a Studio-only feature that costs $300. The creator connects ChatGPT to free Resolve through an open-source MCP server, a connector he compares to a physical dongle bridging two incompatible ports, then scopes the agent's file access to one project folder and hands it a clip with plan-first instructions: cut a highlight reel, caption it, keep it 16x9 for YouTube. ChatGPT proposes a three-tier permission plan, and picking the middle tier, core control plus basic transcription, is the safer default over the most capable option. The whole build, from installing the MCP server to rendering a captioned highlight, cost about 95 cents on the $20-a-month ChatGPT plan.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:00
01 · ChatGPT cuts a 2-hour stream into clips
Cold open showing the finished result: ChatGPT captioning a highlight cut of a livestream in DaVinci Resolve, with the promise that free Resolve normally can't caption at all.
01:00 – 02:22
02 · What an MCP is (the dongle demo)
Uses a physical charging dongle as a prop to explain that an MCP server is the connector that lets an AI agent and DaVinci Resolve talk to each other.
02:22 – 03:28
03 · Setup: requirements + the folder
States the only requirements (ChatGPT and DaVinci Resolve, free version is fine) and shows creating a dedicated project folder for the agent to work in.
03:28 – 04:31
04 · The two skills (in the description)
Opens the Codex app, creates a new local project named 'Assistant AI editor,' and points to the two packaged skills linked in the video description.
04:31 – 05:24
05 · Install the Superpowers plugin
Shows installing the Superpowers plugin from the Codex settings/plugins menu, described as one of his most-used skills.
05:24 – 06:20
06 · The setup prompt
Gives the agent the DaVinci Resolve MCP GitHub repo and asks it to read through the page and help set it up, first in a brainstorm/plan mode that blocks it from taking action.
06:20 – 08:27
07 · The plan: three tiers, pick the middle
Walks through the agent's proposed three-tier permission plan (core control, core plus analysis, full complexity) and picks the middle tier since transcription is already covered by FFmpeg.
08:27 – 10:10
08 · First edit: one video, one instruction
Pastes in an Alex Hormozi livestream clip and gives one instruction: cut the first 20 minutes into a coherent 3-4 minute highlight, 16x9 for YouTube, clean edit plus captions, then approves the resulting plan.
10:10 – 11:41
09 · What it does while you wait
Steps away and lets the agent run in the background, transcribing every clip and picking out the good parts unattended for over 25 minutes.
11:41 – 12:54
10 · The pipeline runs on its own
The agent restarts DaVinci Resolve to refresh the connection, then returns with a rendered highlight cut inside the timeline, all on the free version.
12:54 – 13:34
11 · What it cost (70% to 51%)
Shows the ChatGPT usage meter dropping from 70% to 51% remaining for the whole build, roughly 95 cents on the $20 monthly plan.
13:34 – 13:44
12 · Run it yourself
Points viewers to the two skills linked in the description and pinned comment: install them, run 'Resolve MCP Setup' then 'Resolve Video Pilot.'
Lines worth screenshotting.
- Free DaVinci Resolve cannot transcribe or caption audio, that feature is locked behind the $300 Studio license.
- An MCP server acts like a dongle: it's the missing connector between an AI agent and desktop software that was never built to talk to it.
- Connecting free DaVinci Resolve to ChatGPT lets it transcribe and caption clips without ever paying for the Studio upgrade.
- MCP setups often offer multiple permission tiers, and picking the middle one instead of the most capable one is the safer default on a first install.
- Restricting an agent's file-system access to a single project folder keeps it from touching anything outside the intended workspace.
- Approving an AI agent's plan before it executes works like approving a contractor's quote: nothing moves until you say yes.
- $20 a month went noticeably further on ChatGPT than the same $20 on Claude for this specific editing workflow, even though Claude produces better motion graphics.
- The entire build, installing the MCP server through rendering a captioned highlight, cost about 95 cents in AI usage.
- Once the MCP server is installed, the same setup works on both a single short clip and a full two-hour livestream recording.
- The agent still won't start cutting on its own, a human has to review and approve the plan first.
How to get paid software features from a free tool using an AI agent
AI Agent Workarounds
An MCP server can bridge an AI agent into desktop software that never shipped an AI integration, letting a free app do things its paid tier normally gates behind a license fee.
02What an MCP is (the dongle demo)
- An MCP server is the adapter that lets an AI agent talk to software that was never built to accept AI commands, the same way a physical dongle bridges two incompatible ports.
- The same MCP-based method works with either ChatGPT or Claude, the choice of model doesn't change whether the underlying app can be connected.
03Setup: requirements + the folder
- Setting up an AI-controlled workflow starts with three basic pieces: pick one folder for the agent to work in, approve its plan, then hand it a file.
- You don't need the paid tier of a tool to use this kind of workaround, the free version is the entire point.
04The two skills (in the description)
- Packaging a repeatable AI workflow as a named 'skill' makes it something you can run again later instead of re-explaining the setup from scratch.
- A skill built on one OS can often be adapted by the agent itself for a different OS, so a Windows-built skill isn't necessarily locked to Windows.
05Install the Superpowers plugin
- Extending an AI coding tool with a plugin is usually a few clicks in a settings menu, not a separate install process.
06The setup prompt
- Giving the agent a specific reference (a GitHub repo, a doc) plus a clear goal is enough of a prompt to start a real setup, no need to write out the technical steps yourself.
- Switching an agent into a read-only planning mode before it touches anything lets you review the approach without risking an unwanted action.
07The plan: three tiers, pick the middle
- When an integration offers multiple permission tiers, the most capable one isn't automatically the right pick, more capability also means more that can break.
- Choosing a tool you already have covered over a redundant added capability avoids adding complexity and risk for no real gain.
- Starting with less scope than you think you need is a better default for a first-time install than starting with everything enabled.
08First edit: one video, one instruction
- A single plain-language instruction, length, platform, aspect ratio, and edit style, can be enough direction for an agent to produce a full working plan.
- Approving an agent's plan before execution works like approving a contractor's quote: you say yes or no once, and nothing changes until you do.
09What it does while you wait
- Once a plan is approved, an agent can run unattended in the background for extended periods without a person watching every step.
10The pipeline runs on its own
- The output from a workaround can be a fully finished result, not just a partial stand-in for the feature you didn't pay for.
11What it cost (70% to 51%)
- A full end-to-end build, from installing an integration to producing a finished output, can cost under a dollar in AI usage on a standard flat-rate plan.
- Usage limits on flat-rate AI plans are a real cost constraint worth tracking, a single ambitious build can eat a meaningful share of a week's allowance.
Terms worth knowing.
- MCP
- Model Context Protocol, a connector that lets an AI agent like ChatGPT or Claude read and control a piece of software that wasn't originally built to accept AI input.
- Superpowers
- A plugin for the ChatGPT/Codex app that extends what the coding agent can do, turned on from the app's plugins settings page.
- File-system boundary
- A permission setting that limits which folders an AI agent is allowed to read or modify, keeping it scoped to one project instead of the whole computer.
Things they pointed at.
04:31toolSuperpowers plugin
03:40productChatGPT Plus
Lines you could clip.
02:10
“This dongle here is basically an MCP.”
06:30
“It's one car, so in three trims, the top one isn't a better car. It basically has more stuff that can break.”
08:00
“That's the quote your contractor hands you before anyone knocks down a wall. You read it, you say yes or no, and nothing moves until you do.”
18:13
“You basically have until you're 30 to be considered young. The moment I turned 30, I was just a white dude.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
What you're seeing on screen right now is ChatGPT cutting up a two hour live stream on DaVinci Resolve into a one and a half minute video. And then it's captioning it. So if you've ever tried to caption or transcribe your clips without the DaVinci Resolve Studio version, which is $300, you know that it doesn't work.
So in this video, I'm going to show you how to cut and caption videos using ChatGPT and how to set up your own AI assistant editor. And keep in mind. You can use the same method with Claude as well.
So right here, the Tantra guy, the free version worked. I just let ChatGPT figure it out. Thank you so much.
So shout out to you, the Tantra guy. You're the reason why I am making this video since you confirmed it works. So this method requires what we call an MCP.
Currently, there's no natural way for DaVinci Resolve to allow you to connect your agent into... DaVinci so that you can use your agent and interact and even control DaVinci Resolve. So this is what that video solves.
But for the free version, what is an MCP? Let me explain. So right here I have my map.
We have our charger and these are type C ports. What we want to do is I want to connect this USB port to my MacBook, but. Look at this.
It doesn't connect. And so I want you to imagine that DaVinci Resolve, that this MacBook is DaVinci Resolve, and this CHPT. Naturally, it doesn't connect well with each other when we try to plug it in.
But what if we had a dongle or what we call an MCP? . It provides us a connection where we can plug in.
So we take this here. And connect it to our dongle.
And just like that. This dongle acts like an MCP. I am on my PC.
And this PC does not have a DaVinci Resolve Studio account. It is actually the free one. This is a fresh copy of DaVinci Resolve.
Yeah. See? Right here while it's loading.
There is no studio version on it. It's really just three things. You put it at one folder, you approve its plan, and then you give it a video and let it edit.
That's it. So in order for you, whenever you first open up ChatGPT, the requirements are you need to have ChatGPT and DaVinci Resolve. You can have the free version.
You'll be on this page right here. This next part might look like a lot. I'm just making a folder and telling it where to work.
That's pretty much it. I'm not going to show you how to download and install ChatGPT. You are on your own for that one.
So you're going to go up here on the top left corner. You're going to click on Codex. And right here where it says choose a project, go ahead and choose a project.
I already have my operator OS, but let's say that you've never opened this before. You haven't set anything up. You can go here, add folders Codex can run, or you can just name it Assistant AI.
editor. And this becomes local only. Anytime you want to go or do anything with your AI editor, you're going to click in here, click into your folder, or you can click over here as well.
So I would just keep the options like this for now, assistant AI editor, keep it as local and master branch. What I figured out here, I turned it into two skills. They're in the description.
I built these on Windows, but your agent should be able to adapt them for Mac. You can follow along with the rest of this video and set it up yourself or just run the skills either way. So we have ChatGPT here.
Honestly, I did some tests and I found that ChatGPT is probably the most efficient AI assistant editor you can use. And just to show you, I am currently on the $20 plan, $20 ChatGPT plan. It gets really far.
$20 with ChatGPT is so different from $20 with Claude. Claude has the best front -end design, so that means if you want to create motion graphics and animations, Claude is the better model to do that with. But if you just want to get editing done and you're okay with whatever motion graphic and animations that ChatGPT creates for you, then it's the more efficient and conservative out of the two.
And one thing I forgot to mention inside of the video is how to install superpowers, because this is one of my most used skill. In order to install superpowers, you want to go here, down here. So we have the whole Codex app right here.
You can go here, click on your name, go to settings, and head over to plugins.
In plugins, you'll see browse directory and add. Click on browse directory and you'll type in superpower.
You should see superpowers right here on the screen. You'll click on it. On the top right of your screen, you will see either a switch or button to turn it on.
This is how you enable superpowers. I have a repository that I want you to help me set up. It is to connect Codex with DaVinci Resolve MCP.
But with the DaVinci Resolve MCP, I only have the free version of DaVinci Resolve. So I want you to read through this GitHub page that I'm going to copy and paste in and help me set it up. Use Brainstorm.
didn't really want it to send but that's okay i'm gonna do this and then we're gonna type slash plan mode so that it does not do anything i read through each of these settings and basically number one it gives us the core control which is the main components that we need in order to use ChatGPT as our AI assistant editor.
Where number two, it's the core plus analysis. This one, it sounds pretty good actually. Allows us to do media analysis, such as transcriptions.
That one's really good, but we already have FFmpeg. So with more complexity and risk. So what we're going to do is we're going to do the core plus analysis.
It's one car, so in three trims, the top one isn't a better car. It basically has more stuff that can break. So I went with the middle.
I would essentially recommend that if you want some crazy stuff, honestly, it's better for you to start with less than to start with more, especially if this is your first time installing it. We're going to go with FFN transcription, basically the middle option. Don't try to do everything all at once because you think you can, you'll just end up overwhelming yourself.
This has worked for me pretty well, picking it up and learning it as a beginner. I started with using Cloud Cowork, and that really taught me the ropes on how to use these agents with guardrails. So this specifically, I only want DaVinci Resolve to be able to work.
inside of our project. I'm currently inside of my operator OS. It's basically the folder that I've created that I do all my work in.
This is from my main computer or my main source. So it's asking us the file system boundaries should resolve MCP receive. We're going to do operator OS only.
Let's go ahead. I'm going to review over this plan real quick to make sure it looks good. Yeah, everything looks good.
Let's go ahead and approve this plan now. And . You read it, you say yes or no, and nothing moves until you do.
So that was section one of the design. Now we're on section two. We went ahead and improved section two.
And I think in a moment here, we're going to go over to section three. Here it is, section three. And for this specific test, I'm using an Alex Hermosi live stream that he did recently.
So right here, we wanted to do something with this. Video. I'm going to paste in a video and I am going to say for this video, I want you to cut out the first 10 minutes of it and watch through it too.
Actually, let's change this to the first 20 minutes. Clip it up. Honestly, this part doesn't really need the DaVinci Resolve MCP, but we just want to see it.
with and turn the highlight clip short three to four minute we can go with a coherent mini story we don't need the montage and we don't want condensed cut it's going to be for youtube 16x9 we are not a short form pro channel although this can do short form we are not sure from pro clean edit plus captions yeah let's just do clean edits and captions that's typically that's that's pretty typical of highlight channels anyways so our so this is going to be the final plan right here and once this finished writing the plan i am going to swap this over to i mean not so luna 5 .6 all right let's implement the plan swap our model 5 .6 luna on let's do high all right and from there we just sit back relax I'll put up a usage counter so that you can see what this ends up costing.
But for now, I'm going to step away and let it run. Just so you can see how much token this takes. And I'm going to stop my camera recording.
And this is just going to run in background. If something changes during that time, I'll resume it. But other than that, I'll just keep this screen recording going.
Let's see how long this session has been running. So this session has been running for 26 minutes so far. Right now is transcribing every clip and picking out the good parts and it looks like it's frozen for a moment But it's working in the background All right looks like the transcript is finished now and I'm currently allowing codecs to Take over my PC in order for it to set everything up.
It's restarting DaVinci Resolve that the Connection will renew most likely Meanwhile the free version it built itself a side door. That's why it resets. Okay So it works You guys are currently seeing shrink this thumbnail All right, shrink this timeline a little bit.
So we should see this getting rendered in a moment here. This version two over here. So that live stream clip I gave it earlier, this is what it came back with.
I'll give you a secret that I only found out later is that you basically have until you're 30 to be considered young. The moment I turned 30, I was just a white dude. Before that, I was a young white dude.
That's the finished cut. Now I want to go back to the captions part. because on the free version of resolve that would not even be possible analyze okay that's good build and render the three to four minute pilot then burn captions run final media safety connect and connection checks all right we're almost done pretty much setting up the free davinci resolve on our pc so anytime we want to use davinci resolve on here at least if we want to use our ai assistant editor we don't need the studio version and you can choose from there if you want to that's how you would customize your captions with different fonts text colors design something that i do have to mention is that if you didn't know davinci resolve has a few features that you cannot do And one of them is actually audio transcription, I think.
Yeah, you cannot transcribe regular video clips if you do not have the studio version, as you can see here. And this also shows that I don't have the studio version. So this is essentially us getting around that fix.
So now we know that. This feature, it works for the free version of DaVinci Resolve. Initially, when we were doing this build, we were at 70%.
Now we're at 51%, so effectively 19%. Now that's roughly a fifth of my week. On a $20 plan, that's like 95 cents of our usage limit was from installing this Resolve free MCP.
Well, I hope this video has helped you out. If you have DaVinci Resolve or you've used DaVinci Resolve or you have the free version, the skills are linked in the description and pinned comment if you want to set this up for yourself. Now with the skills, all you have to do is install the skill onto Codex or Cloud, whichever the case is.
It will run through, run the skills, say Resolve MCP Setup. It will go ahead and run that. Then you can do Resolve Video Pilot.
From there, it will run both and you'll be good.
The Hook
The bait, then the rug-pull.
The creator connects free DaVinci Resolve to ChatGPT through an open-source MCP server, then hands it a two-hour livestream and gets back a captioned highlight reel, something the free version of the software isn't supposed to be able to do.
Named ideas worth stealing.
06:30list
The three MCP permission tiers
- Core control (main components needed to use ChatGPT as the editor)
- Core plus analysis (adds media analysis like transcription)
- Full complexity (more capability, more that can break)
ChatGPT proposes three levels of access when setting up the DaVinci Resolve MCP connection, ranging from minimal control to full-featured but riskier.
Steal forany first-time AI-agent integration with a new tool, where the instinct is to grant the most permissions available
How they asked for the click.
VERBAL ASK
13:36link
“the skills are linked in the description and pinned comment if you want to set this up for yourself”
Soft, single-ask resource CTA at the very end pointing to a free kit.com link with the two Codex skills; a Calendly link for a free 1:1 editing chat also sits at the top of the description but isn't spoken on camera.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
