Big Idea

The argument in one line.

An MCP server turns DaVinci Resolve into something an AI coding agent can directly operate, letting it perform a real structural edit unattended, cutting runtime by a third while still leaving the final judgment call to a human.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You create long tutorial or talking-head videos and editing is the biggest bottleneck between filming and publishing.
You already use an AI coding agent like ChatGPT/Codex or Claude and don't mind installing an MCP server to extend it.
You edit in DaVinci Resolve and want AI to take a first pass at trimming dead air, retakes, and filler before you touch the timeline.
You're curious what MCP (Model Context Protocol) actually does and want to see it demonstrated end to end, not just explained.

SKIP IF…

You need a finished, publish-ready edit in one pass. This workflow still ends with a manual review, not a final export.
You don't edit in DaVinci Resolve or use an AI agent that supports MCP servers.

TL;DR

The full version, fast.

Steven Liu installs Samuel Gursky's open-source DaVinci Resolve MCP server, which exposes Resolve's editing API to any AI agent that supports MCP. He confirms ChatGPT (Codex) can see and control Resolve, then gives it a detailed prompt: transcribe the raw footage with Whisper, remove only high-confidence dead air, filler, and repeated or abandoned takes, and build a new versioned timeline without touching music, effects, or transitions. After roughly an hour of unattended work, the AI delivers a new timeline 34% shorter than the 38-minute source, reorders the opening to lead with the finished product instead of the setup, and removes multiple abandoned explanation attempts. The result isn't publish-ready. Liu still recommends a full manual review pass, since the AI's own QA was spot-checked previews rather than a complete watch-through.

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 →

Chapters

Where the time goes.

00:00 – 00:27

01 · The editing bottleneck

Liu states the problem (editing is his biggest bottleneck) and the promise: an AI workflow in DaVinci Resolve that cuts editing time without sacrificing quality.

00:27 – 01:47

02 · Installing DaVinci Resolve MCP

Explains what an MCP server is (the USB-C-for-AI analogy), then finds Samuel Gursky's DaVinci Resolve MCP on GitHub and installs it, dragging the raw tutorial footage into the media pool.

01:47 – 02:16

03 · Verifying ChatGPT has Resolve access

Asks ChatGPT directly whether it has access to the DaVinci MCP; it confirms it can control Resolve for editing, timelines, and media, and can launch Resolve automatically.

02:16 – 03:36

04 · Writing the edit prompt

Sets the model to a higher reasoning tier and pastes a detailed prompt: use the MCP to edit the raw tutorial into a natural YouTube video, transcribe with a fast Whisper model, and remove only high-confidence dead air, filler, and abandoned takes.

03:36 – 04:06

05 · Touring what the MCP can do

While the AI works, Liu scrolls the DaVinci Resolve MCP GitHub repo showing its full capability list: project control, timeline editing, review annotations, color grading, and Fusion.

04:06 – 04:35

06 · AI edits unattended

Notes that the AI is never perfect and the process is iterative, not a single prompt and done. After about an hour and nine minutes, the edit is ready to review.

04:35 – 05:27

07 · The result: 34% shorter, reordered opening

The new timeline runs 25:16 versus the original 38:16, a 34% reduction. The AI also reordered the footage, moving the finished-product demo from the end of the raw recording to the front as the new opener.

05:27 – 06:22

08 · Cutting repeats and abandoned takes

Reviews the AI's cut list and self-audit transcript, including specific timestamps where it removed a corrected mistake and an abandoned explanation attempt, keeping only the final clean version.

06:22 – 07:21

09 · Before and after: the OpenAI setup clip

Plays the original raw clip where Liu stumbles through explaining an OpenAI setup step, then plays the AI's cut version side by side to show what was kept and what was removed.

07:21 – 07:29

10 · Sign-off

Asks viewers to comment their verdict on AI editing and to like, subscribe, and follow up if they want more on this workflow.

Atomic Insights

Lines worth screenshotting.

An MCP server exposes an application's own API to any AI agent that supports the protocol, so DaVinci Resolve doesn't need a custom integration for each AI tool.
The AI reduced a 38-minute raw tutorial to roughly 25 minutes, a 34% cut, using a single detailed prompt and about an hour of unattended work.
The AI reordered the video's structure, not just trimmed it: it moved a demo of the finished product from the end of the raw footage to the very beginning as the new opener.
Giving the AI a narrow, high-confidence cutting mandate (dead air, filler words at safe boundaries, abandoned takes) rather than 'edit this however you want' is what kept the cut faithful to the source material.
The prompt explicitly forbade adding music, effects, captions, zooms, or transitions, keeping the AI's job scoped to structural trimming rather than stylistic decisions.
The workflow includes a built-in self-check: after the structural edit, the AI re-transcribes its own new timeline and audits it for broken sentence joins, mismatched audio, and cut errors before calling the job done.
The AI can identify and remove 'abandoned attempts,' cases where the creator started an explanation, made a mistake, and restarted, keeping only the corrected final version.
Even after an automated edit and self-audit, the creator still recommends a full manual playback review, because the AI's own verification used spot-checked previews and screenshots rather than watching the entire new cut.
The edit is not a single prompt and done: the creator frames it explicitly as an iterative process, expecting to give follow-up instructions after reviewing the first pass.

Takeaway

AI can structurally edit a video, but it still needs a human review pass.

WHAT TO LEARN

An AI agent with direct API access to your editor can trim dead air, cut repeated takes, and even reorder your structure, as long as you scope its cutting mandate tightly and still watch the result before publishing.

02Installing DaVinci Resolve MCP

An MCP server exposes an application's own API to any AI agent that supports the protocol, so a tool like DaVinci Resolve doesn't need a custom integration for each AI you use.

03Verifying ChatGPT has Resolve access

Confirm the AI actually has the access you think it has before handing it a real task; a one-line check ('do you have access to X?') catches setup problems early.

04Writing the edit prompt

A narrow, high-confidence cutting mandate (dead air, safe filler, abandoned takes) kept the AI's edit faithful to the source, rather than an open-ended 'edit however you want' instruction.
Explicitly forbidding stylistic additions (music, effects, captions, zooms, transitions) in the prompt kept the AI's job scoped to structural trimming instead of creative decisions.

07The result: 34% shorter, reordered opening

The AI reduced a 38-minute raw recording by 34% and reordered the opening to lead with the finished product, a genuine editorial judgment call, not just a trim.
Building a re-transcribe-and-audit step into the prompt gave the AI a way to catch its own broken cuts and mismatched audio before calling the job finished.

08Cutting repeats and abandoned takes

The AI can distinguish an abandoned, restarted explanation from the corrected version that followed it, and keep only the clean take.

09Before and after: the OpenAI setup clip

Treat an AI edit as an iterative first pass, not a finished product: this workflow's own self-audit used spot-checked previews rather than a full playback, so a manual review is still required before publishing.

Glossary

Terms worth knowing.

MCP (Model Context Protocol): A standard way for an AI agent to connect to and control an application without a custom integration. The video describes it as one common connector, comparable to USB-C, that lets any MCP-compatible AI talk to an app like DaVinci Resolve.
DaVinci Resolve MCP: An open-source MCP server (by developer Samuel Gursky) that exposes DaVinci Resolve's editing API, letting an AI agent perform actions like listing projects, editing timelines, adding overlays, and adjusting color grades.
Structural edit: In this workflow, the first editing pass where the AI cuts dead air, filler, and repeated takes and reorders footage, without adding any stylistic elements like music or captions.

Resources

Things they pointed at.

00:59toolDaVinci Resolve MCP (Samuel Gursky, GitHub) ↗

01:18toolChatGPT / Codex

02:45toolFast Whisper model

00:00linkhundredacretech.com (Liu's app-development business) ↗

00:00linkcal.com/stevenliu (discovery call booking) ↗

Quotables

Lines you could clip.

00:51

“Think of MCP as a USB-C for AI, one standard connector that works with many different devices and softwares.”

clean, self-contained analogy that explains the whole premise→ TikTok hook↗ Tweet quote

04:00

“Of course, do you know that AI is never perfect? So there may be changes required. It's not a single prompt and it'll be done. There will always be iteration after iteration.”

honest expectation-setting about AI editing→ newsletter pull-quote↗ Tweet quote

08:03

“It's better placed at the front of the tutorial so we can actually look at the raw transcript that it has shown.”

shows the AI made a real editorial judgment call, not just a trim→ IG reel cold open↗ Tweet quote

The Script

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.

Have you ever spent hours editing a single video or paid hundreds of dollars to have someone else do it? For me, editing has become the biggest bottleneck for creating YouTube content. I started building an AI -powered workflow with DaVinci Resolve that cuts my editing down dramatically without sacrificing quality.

In this video, I'll show you exactly how I use AI to handle repetitive work while I stay in control of the final edit. If you create YouTube videos, tutorials, this workflow could save you hours. Let's get started.

First thing we need is DaVinci Resolve MCP. So you might not have used MCP servers before. What MCP does is it allows AI to securely talk to other apps using one common knowledge.

So for example, instead of teaching ChatGPT how to access DaVinci, DaVinci can expose an MCP server and its server instructions, and any AI that supports MCP can connect to that MCP server without using custom integration. So as an analogy, think of MCP as a USB -C for AI, one standard connector that works with many different devices and in softwares in this case.

So if you go to GitHub and search Samuel Gursky's DaVinci Resolve MCP, or simply Google it, there's a set of instructions to install it. It's quite simple, or you can get AI to install it for you. Once you have your MCP set up, go to your preferred AI agent, and here I'm using Codex, or now called ChachiBT.

To verify that has successfully installed, you can go to settings and then you go to plugins and in MCPs, there should be DaVinci Resolve MCP right here. Secondly, I have a full unedited raw video of a tutorial where I'm explaining on building an AI knowledge base. Here, I'm not gonna do anything.

I'm just gonna drag in the assets into the media pool. What I'm gonna do now is validate whether ChatGPT has access to DaVinci MCP. So I'm gonna briefly ask it.

hey do you have access to the davinci mcp and i'm going to leave my model as so high of course depending on your use case or your budget you might be able to go extra high or even ultra or light or even change a different model to terra or luna so here chat gp has responded i have access to the davinci mcp and can control resolve for editing etc etc it can also launch resolve automatically if needed i already have resolve launched but let's paste a prompt that i want to allow this video to be edited so what i'll do here is i'm going to change the models to so extra high of course the higher the model the better it will be and i have pasted this prompt that i have created already but you can change the prompt to however you want the ai to edit the video so what i have here is basically a prompt use the davinci mcp to edit the spoken tutorial ai edits which is what the project is called on davinci right now into a natural youtube video preserve or consider

existing timelines and create new version timelines what I want to do here is basically transcribe the video using a fast whisper model to a transcript and then use stronger model on short whenever speech is ambiguous what I want to do here is edit unnecessary material the end goal is to trim all the unnecessary cuts where I am in silent or if I'm repeating myself because I made a mistake and later on I corrected it and things like that.

And the end goal of this is to have an edited video where we are removing a lot of the mistakes and silences and necessary information to ensure this first iteration of the edit is good for my review. So let's submit this and let's see what AI will do now. While that's working, we can go down, go back to the GitHub DaVinci MCPU.

repository and show you what it can do so if you scroll down it can actually do heaps of things for example it can list all projects it can create fusion text plus overlay over selected clips etc the core capability is anything almost davinci and davinci api can do or you can app and project you can do project control you can do fusion even color grading AI can basically tap into this MCP to do all these things.

But when you come back, whether after your nap or after you go do something else, it has made these changes for you to review. Of course, do you know that AI is never perfect? So there may be changes required.

It's not a single prompt and it'll be done. There will always be iteration after iteration. So after ChatGP has worked for one hour, eight minutes and 56 seconds, There is a edited version in DaVinci Resolve.

So if we jump into DaVinci Resolve, you can see here it has placed all the cuts into the timeline and had made all the changes inside. What's done is the original timeline was actually 38 minutes. It was reduced by 13 minutes, which is 34%.

What it actually did was it actually changed the timeline a little where it reordered the end product opening. at the very beginning because i filmed the video where they i showcased the end product at the end of the video and then it figured out that that was the end product and it's better placed at the front of the tutorial so we can actually look at the raw transcript that it has shown so this is the source transcript i start with hey everybody hope you're doing well in this video we're going to build an ai knowledge base and explain what is knowledge base but then in this new version It actually changed it where this is what we're building at the end.

So the end customer lands on knowledge base and so on. So that's how it was smart enough to figure out, hey, we need to have this interesting opener at the front. So the viewers like yourself go straight to understanding what they're getting out of it instead of having an explainer of what we'll be building.

And then it made all these cuts about, like for example, if I repeated something, it made the cuts and then, selected the right one which is the latter one all the time i got ai to identify example where there were abandoned attempts so in the source it said around 5 50 5 minute 54 to 606 there is abandoned attempts where i had two abandoned attempts to explain why open ai is needed and then it removed it for me so if we go to the original video let's watch the original video where it says this is the 554 mark if you go to platformopenai .com this allows you to create a can this allows you to leverage open ai services on a paid on a you this will allow you so you can see i made a lot of mistakes there and let's see what ai cut it to so this is the cut that ai kept so let's hear it and see how does it differ secondly we'll need open ai so this will allow you to use open ai's services on a paid basis so once you've set it up create a new project and then so this is quite good because actually removed the initial part but it kept some context about requiring open ai and then jumped to why we needed open ai while the source was kind of just like uh a lot of broken words that didn't make sense at all and they had to do a retake if you really enjoyed this video comment below and let me know if you want to

hear more about how you can use AI to edit videos. And as always, remember to comment, like, subscribe and see you next time. Bye.

The Hook

The bait, then the rug-pull.

Steven Liu opens with the pain every solo creator knows: editing eats more time than filming. His fix is to give an AI coding agent direct control of DaVinci Resolve through an MCP server, then watch it cut a real 38-minute tutorial on its own.

Frameworks

Named ideas worth stealing.

02:17list

The AI structural-edit prompt

State the goal: edit the spoken tutorial into a natural video, preserving existing timelines by creating a new versioned one
Transcribe first with a fast model, escalate to a stronger model only when speech is ambiguous
Remove only high-confidence unnecessary material: dead air, safe filler words, abandoned takes, repeated explanations
Explicitly forbid stylistic changes: no added music, effects, captions, zooms, or transitions
Re-transcribe the new timeline and run a second semantic audit for broken cuts and mismatched audio before finishing
Report back: new runtime, reduction percentage, structural changes made, and anything requiring human review

The prompt Liu wrote to keep the AI's edit scoped to safe, structural cuts rather than open-ended stylistic changes.

Steal forany first-pass AI edit of long-form talking-head or tutorial footage

CTA Breakdown

How they asked for the click.

VERBAL ASK

07:22subscribe

“If you really enjoyed this video, comment below and let me know if you want to hear more about how you can use AI to edit videos. And as always, remember to comment, like, subscribe and see you next time.”

Soft, single ask at the very end, tied to a direct follow-up question rather than a hard sell.

MENTIONED ON CAMERA

00:59toolDaVinci Resolve MCP (Samuel Gursky, GitHub) ↗

00:00linkhundredacretech.com (Liu's app-development business) ↗

00:00linkcal.com/stevenliu (discovery call booking) ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.