DaVinci Resolve 21.1's Native MCP Server, Tested With Claude and ChatGPT
Team 2 Films runs Claude through four real edit jobs, project setup, a rough assembly cut, TIFF-cached rendering, and color grading, to see what Resolve's new AI agent integration can and can't do.
Posted
4 days ago
Duration
Format
Demo
educational
Views
62.5K
1.6K likes
57 · 43
Big Idea
The argument in one line.
DaVinci Resolve 21.1 ships a native MCP server that lets Claude or ChatGPT call the real editing API directly, so an agent can organize footage, sync audio, assemble a rough cut, cache renders, cut vertical shorts, and grade color largely unattended, capped only by whatever functions Blackmagic has exposed so far.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You edit in DaVinci Resolve Studio and want to know exactly what its new native AI agent integration can and can't do before trying it on a real project.
You're weighing whether an AI agent could handle procedural editing tasks, syncing audio, organizing bins, cutting vertical shorts, so you can spend your own time on the creative decisions.
You manage multi-camera or multi-format footage, mixed ARRI, RED, and scanned film, and want to see how an agent handles timecode sync and per-camera color workflows.
SKIP IF…
You don't use the Studio (paid) version of DaVinci Resolve. This feature isn't available in the free edition.
You're looking for a finished, ready-to-publish edit. Every output shown here, the rough cut, the vertical shorts, still needed the editor's finishing pass.
TL;DR
The full version, fast.
DaVinci Resolve 21.1 ships a native MCP server, so Claude or ChatGPT can call Resolve's real API directly instead of clicking through the interface, which means its abilities are capped by what functions Blackmagic has exposed, not by the model. Team 2 Films runs it through project setup, audio sync, a rough assembly cut, TIFF-cached rendering, vertical-short extraction, and color grading, watching it patch around missing API functions, like downloading FFmpeg itself, or literally moving the mouse to build grading nodes, rather than fail outright. The verdict: it's a capable assistant editor for procedural, time-consuming steps, syncing, organizing, rough-cutting, but it can't watch footage or make the emotional judgment calls creative editing depends on, so every output still needs a human finishing pass.
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.
Resolve 21.1 adds a native MCP server. The agent calls Resolve's real API functions directly rather than operating the GUI, so it's limited to whatever functions that API currently exposes.
01:06 – 02:07
02 · Setup
Requirements: the Studio version of Resolve, an AI agent that supports the Anthropic MCP protocol, a Claude Pro subscription and desktop app, and Resolve 21.1. Turned on under File > Set Up AI Assistance.
02:07 – 04:23
03 · Demo 1
Claude creates a new project ("PhotoChem") and, given only a partial folder name, locates and imports the matching media structure from a network share. It then hits its first API limitation: Resolve's 'update timecode from audio track' function isn't exposed.
04:23 – 05:23
04 · Audiio
Sponsor break for Audiio's royalty-free music library: the Elements 2.0 stems/instrumentals view and the Hanz AI natural-language search, plus a $60/year discount code.
05:23 – 08:33
05 · Demo 1 Continued
Claude routes around the missing timecode function by downloading FFmpeg itself, reads each clip's timecode, and writes it back through a function Resolve does expose. It removes the noisy audio-timecode channel, syncs dialogue to the right camera bins by color, gives a written report of what it changed, then detects scene cuts on scanned film reels and splits them into individual shots, correctly ignoring one-frame flash cuts. It's then asked for a full rough 15-minute assembly, which it builds overnight from auto-transcribed dialogue.
08:33 – 10:58
06 · Demo 2
Instead of step-by-step prompts, Claude gets one long instruction covering an entire render pipeline: cache the timeline as a TIFF sequence in small chunks to an external Iodine drive, re-import it, then render the final H.265 master. It builds multiple render jobs (not one giant job) and can recover on its own if Resolve crashes mid-run.
10:58 – 13:11
07 · Demo 3
Asked to pull 20 roughly 60-second vertical-short candidates out of a 50-minute horizontal documentary, Claude builds 20 duplicate vertical (2160x3840) timelines with 15-second handles and in/out markers, and triggers Smart Reframe on each. It can't clear the blank gap left at the head of each duplicated timeline, a current API gap, and it's choosing purely from the transcript since it can't watch the footage.
13:11 – 14:36
08 · Demo 4
On a timeline mixing ARRI Log, RED Log, and scanned Cineon Log film, Claude is asked to build four serial nodes per clip (Primary, LUT, Secondary, Look), set the project to DaVinci YRGB with Rec.709 Gamma 2.4 output, and apply the correct LUT per camera. Node creation and labeling isn't in the API yet, so it takes over mouse control to build the nodes, then finishes the rest, including correctly sorting clips by camera, through the API.
14:36 – 16:48
09 · Should you be worried?
The host's take: AI is not ready to replace editors. It reproduces patterns rather than drawing on lived experience, so it fits as an assistant editor for procedural work, not for the emotional judgment creative editing depends on.
16:48 – 17:54
10 · The Future?
Every capability shown is bounded by what Blackmagic has exposed in Resolve's API (e.g. no access yet to IntelliSearch visual understanding) and by the AI agent's own compute, not by the model's intelligence. Both are expected to expand.
17:54 – 18:34
11 · Suggestions
Because Resolve is locked while a prompt is processing, the host suggests running prompts on a second, low-power machine (their example: a Mac Mini M4, since the heavy compute happens in the cloud) or queuing up overnight runs.
18:34 – 18:59
12 · Outro
Standard like/subscribe close.
Atomic Insights
Lines worth screenshotting.
DaVinci Resolve 21.1's AI agent doesn't click through the interface at all, it calls the same API functions a Python script would, so it's capped by whatever functions Blackmagic has exposed, not by the AI model's intelligence.
Given only a partial folder name, the agent located and imported an entire network-share media structure without a full file path.
When Resolve's API had no function to read timecode from an audio track, the agent downloaded FFmpeg itself and used it to feed the missing timecode data into a function Resolve did expose.
The agent generated a written report of exactly which audio channel it removed and how it synced dialogue, a form of documentation manual editing never produces on its own.
Told to build a rough 15-minute assembly cut, Resolve first auto-transcribes every clip so the agent can use spoken content, not visuals, to decide scene order.
A single detailed prompt covering caching, rendering, and file placement kept the agent working unattended for hours, including recovering on its own after a Resolve crash.
Asked for 20 vertical-short candidates from a 50-minute cut, the agent chose them from the transcript alone, since it cannot watch the footage it edits.
The agent could trigger Smart Reframe and build 20 duplicate vertical timelines with handles, but couldn't clear a leftover blank gap at the head of a duplicated timeline, a current API gap.
For color grading, the API had no function to create or label nodes, so the agent fell back to literally moving the mouse for that one step, then finished the rest through the API.
The host's verdict is that this is an assistant editor for procedural work, not a replacement, because it can't draw on lived experience to make the emotional decisions creative editing runs on.
Because Resolve locks up while a prompt is processing, the suggested workaround is running prompts on a second, low-power machine like a Mac Mini M4, or queuing requests overnight.
Takeaway
What Resolve's AI agent can and can't do for you yet
WHAT TO LEARN
An AI agent wired into Resolve's real API is only as capable as the functions Blackmagic has exposed, so it's already strong at procedural, time-consuming steps and still blind to the footage itself.
02Setup
The AI agent needs the Studio version of DaVinci Resolve, Claude's desktop app, and Resolve 21.1 configured under File > Set Up AI Assistance before it can touch a project.
03Demo 1
Given only a partial folder name, the agent located and imported an entire network-share media structure without a full file path, useful when project structure changes.
If a Resolve function isn't exposed in the API yet, like reading timecode straight from an audio track, the agent will simply say so rather than fake it.
05Demo 1 Continued
When an API function is missing, the agent can route around it by combining outside tools, in this case downloading FFmpeg itself to read timecode and feeding it back through the exposed update function.
The agent can produce a written report of what it changed, which audio channel it removed, how it synced dialogue, a paper trail manual editing in Resolve's UI doesn't generate on its own.
06Demo 2
A single detailed prompt covering an entire multi-step workflow, cache the timeline as a TIFF sequence, split it into small render jobs, then output the final H.265, let the agent work unattended for hours and recover on its own if the software crashed.
07Demo 3
Asked to find 20 vertical-short candidates in a 50-minute cut, the agent worked purely from the transcript, not the picture, so treat its selections as a shortlist to review, not a final cut.
It couldn't close the leftover blank gap left behind by duplicating a timeline, a current API gap worth knowing about before you queue up a batch of these.
08Demo 4
For color work, the agent applied LUTs and configured color space through the API, but the API has no function yet for creating or labeling grading nodes, so it fell back to literally controlling the mouse for that one step.
09Should you be worried?
The agent found timecode and cut scenes on its own, but it never watched the raw footage, so any content judgment stays entirely on the editor supervising it.
10The Future?
Every capability shown here is capped by what Blackmagic has exposed in Resolve's API, not by the AI model, so the ceiling moves as Blackmagic adds functions, not as models improve.
11Suggestions
Because Resolve is locked while an agent prompt is processing, running prompts on a second machine or queuing several requests to run overnight keeps your main edit bay free.
Glossary
Terms worth knowing.
MCP (Model Context Protocol)
The open protocol that lets an AI agent connect to and call an application's real functions, in this case DaVinci Resolve's own API, instead of clicking through its interface.
API (Application Programming Interface)
The fixed set of functions a program exposes for other software, or an AI agent, to call directly, bypassing the visual interface entirely.
LUT (Look-Up Table)
A preset color conversion that remaps an image's values, used here to convert log-recorded camera footage into a standard viewing color space.
Log color space
A flat, low-contrast recording profile, such as ARRI Log or RED Log, that preserves maximum dynamic range in the original footage so it can be graded later.
Tentacle Sync
A wireless timecode generator system that stamps separately recorded audio with the same timecode as the camera, so the two can be auto-aligned in post.
Smart Reframe
DaVinci Resolve's automatic reframing tool that repositions a shot's crop within a new aspect ratio to keep the subject in frame.
Timecode
The running clock stamped on video and audio during a shoot, used to line up footage from multiple cameras and separate audio recorders to the same moment.
Serial node
An ordered link in DaVinci Resolve's color-grading node graph, where each node applies one adjustment before passing the image on to the next.
“You can think of this sort of AI as an assistant editor.”
the takeaway framing in one sentence→ newsletter pull-quote↗ 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.
17px
DaVinci Resolve 21 .1 is out and now features a native MCP server for interacting with AI agents. Previously, third -party MCP servers had been created for Resolve. That functionality is now native to Resolve.
Something really important to note, your AI agent isn't interacting with DaVinci Resolve's GUI like you do when you edit. it's using davinci resolve's application programming interface or api for short it's a huge collection of functions that you can control with a scripting language like python well instead of writing scripts your ai agent interacts directly with resolve invoking those functions that means that its functionality is limited by the things you can do using resolves api you can't do everything But new functions keep on getting added with every Resolve release.
And given how popular a genetic interaction is likely to become, we think we can expect Resolve's API to keep on growing. Let me show you how it works.
This feature is only available in the Studio version of DaVinci Resolve. It will work with any AI agent that supports Anthropix MCP protocol. I'm using Claude and I have a Pro subscription.
I already have the cloud desktop app installed. I also have Resolve 21 .1 installed. Head to file, set up AI assistance, and follow any prompts in cloud.
And that's it. It's all installed. A couple of things to note.
Because this is our first time, we will encounter several permission prompts in cloud. Secondly, the speed of responses is dependent on the speed of the AI agent you're using. This can be affected by the cost of your plan, the complexity of your request, or the popularity of the service you're using.
We've sped up responses in the video. I'm going to interact with Claude by typing. It is possible to use voice chat as well.
This is definitely making us think of Star Trek 4. Hello, computer.
We will create a project in Just Media, organize and construct an initial rough edit using nothing. but AI prompts. Please open DaVinci Resolve and create a new project called PhotoChem.
So far, so good. My media is on the share. I can't remember the exact file path.
Let's tell Claude what I do know. There's a folder on the project's network share. It has PhotoChem in the name.
Please import the media in the capture folder, recreating the structure within.
Simple but impressive. Claude found the network share based on a partial folder name and correctly identified the media within it. Next, let's start organizing the media.
Here's where we might hit a limitation of Resolve's API. Let's check something with Claude. ACAM has timecode.
BCAM has audio timecode. Can you access Resolve's update timecode from audio track function?
Okay, as suspected, update timecode is not exposed in Resolve's API. As mentioned, as the popularity of workflows like this increase, we think we can expect features like that to get added. But what can we do in the meantime?
Well, it would be pretty easy for me just to go trigger this function manually myself, the old -fashioned way, with a mouse. It's just in the right -click context menu within the media pool. But I wonder if Claude can think of another way to solve this problem.
Let's go with option two. Cloud is going to download FFmpeg, read the timecode from each clip, and then use Resolve's API to update the timecode of each clip. Resolve's API does have a function for that.
This will take a while. It's definitely quicker to do this in Resolve using the right -click option, but this is a fun demonstration of how an AI agent can solve problems by combining different services and programs into a single workflow. This video is sponsored by Audio where with our discount code you can get 70 % off an entire year of royalty -free music.
We use Audio for all of our music. I'd like to highlight a genuinely useful feature and that is their phone app. The homepage features new artists and playlists.
That's a great place to start if you don't know what you're looking for yet. If you're out and about, when you've got time to kill, instead of listening to the music you already know, you can explore Audio's library and start building ideas for your next edit. If you're looking for something specific, head to search.
You can do all the normal stuff like search through song titles, artists and keywords, or you can hit this button in the middle. This is Hanz AI. It understands the music in Audio's library.
You can either describe the kind of song you're looking for or describe the kind of video you're looking to make. Either way, when you find songs you like, mark them as a favorite. Now when you sit down to start editing, instead of starting from zero, you're already ahead.
You can get an entire year of Audio Pro for just $60 using our discount code in the description. Now that the timecode has been updated, let's remove the audio channel that the timecode was recorded on so we don't hear that awful noise when editing remove the audio channel with the time code on from the b cam very cool and did you notice that claude was able to identify which channel the left or the right had the time code in it next let's sync the audio we recorded dialogue using tentacle track ease they need to get synced to the video Sync the blue and yellow audio to the A and B cam.
Retain the on -camera audio.
It has correctly understood the context that the blue and yellow refer to bins with audio recordings. It's invoked the correct API calls to sync the audio and it's kept the on -camera audio. It's even given me a report.
You don't get this level of detail when working in Resolves UI. So even if this is easy to do within Resolve, you might choose to do this using an AI agent because it's able to analyze what has been done. There's one last thing to do before we start getting creative.
DCAM is shot on film. That means there's one clip per scanned reel. Break it down into individual shots.
In the DCAM bin, there should be one clip for each shot.
It looks like Claude is doing this using the timeline detect scene cuts feature. Again, we have sped this process up. Incredibly, without prompting, it has noted that some of the cuts are very short, just one frame, and therefore are likely frame flashes between cuts.
So it's going to ignore them. And it's also worth noting that it has intelligently combined a selection of functions to detect the scene cuts in a timeline, then place them back in a bin and rename them. It's time to start creating the edit.
In this video, we tour Photokim's lab and learn about the production of 70mm prints. Please can you make a rough 15 -minute edit? Reorder the scenes as appropriate to follow the process.
Where there are clips with matching timecode, please place them on the second video track. All edits should be kept in a bin called 00timelines.
This is going to take some time. Because in order to understand the contents of each shot, Resolve has to first transcribe all the clips. You can see it triggers transcription in the background.
Claude can then use those transcripts to build a narrative. The great thing is that the AI agent will keep on working in the background until the task is done. This would be a great point for me to head out to lunch.
We will jump forward again. Here's Claude's rough assembly edit. It's going to need a lot of work to craft into a finished video, but AI has helped me bypass the time -consuming task of syncing audio and cameras and assembling the story in the correct order.
We use Cloud and Resolve to build a project from scratch, but it's also super helpful when working on projects that have already been progressed. In our previous example, we gave Cloud instructions one at a time. Let's try telling it everything at once and just letting it fly.
When we render long and complex timelines, like our recent Kodak Factory tour or our immersive adventure, it can take 10 to 20 hours. We'll often create a manual cache of the timeline using a TIFF image sequence. That means if we spot minor errors, we only have to re -render the affected frames.
Then encoding the final video takes hours rather than days. Let's see if we can use AI to automate that process. That TIFF cache will require many terabytes of fast data.
I'm going to use this Iodine ProData to store it. So, let's give Claude its marching orders and see how well it does. To make it easier to render my final delivery, I want to cache the entire timeline as a TIFF sequence.
Please render the timeline in small sections in case there are issues. Store the TIFFs in a folder on my Iodine external drive. Re -import the TIFF renders into a bin called Assembly Cache and then place them in the highest video track in the timeline.
When the cache has been created, please render the entire project using the H .265 master preset. Render that final delivery to the iodine drive as well. Please place it in an appropriate folder.
As you can see, it has created multiple render jobs to produce the TIFF sequence. Not just one giant job, and it's correctly identified the drive that I wanted those files to be stored on. What's really cool is that I can leave this to run unattended, and if there are any problems, Cloud can spot them and adjust its plans.
Even if Resolve crashes, it can spot that, reopen, and problem solve. Once finished, this TIFF cache will make it easier to produce final deliveries, and I'm grateful for this help managing the files and processes.
An increasingly important task for editors is to create vertical cutdowns of long -form horizontal content. Can AI assist with that? This is our 50 -minute -long Kodak factory tour.
Let's see what it can do. I'd like to extract vertical shorts from the currently selected timeline. Each should be around 60 seconds long and feature an engaging hook.
If needed, you can make edits to help each segment flow. Make 20 suggestions. Create a timeline for each.
The timeline should contain the original media. Duplicate the original timeline so that the vertical shorts have the same audio track layout and processing. Please include 15 second handles before and after the short to allow me to finesse the edit.
Add a duration marker that shows the portion of the timeline you are suggesting at the short. Don't forget to change the settings for these timelines to make them vertical, 2160 by 3840. Change the scaling so that the clips fill the screen and trigger Smart Reframe for each clip.
The timeline should be stored in a subfolder of the 00 timelines bin. This job will probably take a while. As I've already had lunch, I'll go and have a tea break.
And this is what I'll find when I get back. There are 20 timelines in the bin as requested with handles. Claude has given them names that reflect their content.
The portion that is actually being suggested for the short is marked with in and out points. And smart reframe has been run. There is one small issue.
Claude wasn't able to remove the empty gap at the beginning of the edit. That's there because this timeline was duplicated and the stuff we didn't want was removed. The inability to remove those blank gaps is an API limitation.
But as said before, we expect to see this improve in the future. These are no way ready for delivery. They need a lot of work.
But this has saved me a lot of time. I can watch through these 20 edits, choose the ones with the most potential, and carry on refining them myself. Remember, Claude cannot watch the footage.
It's making these suggestions. based purely on transcribed dialogue.
Finally, let's see if AI can help me color grade. I'm in a timeline with media from different cameras. This timeline is filled with media from an ARRI camera in ARRI Log 3, a RED camera in RED Log, and some media shot on film and scanned in Cineon Log.
Create four serial nodes on each clip and name them. Primary, LUT, Secondary, Look. configure the project to use DaVinci YRGB, set the output color space to Rec .709 Gamma 2 .4, apply the correct LUT to each clip on the appropriate node.
While the Resolve MCP server does not allow Claude to interact with Resolve's GUI, actually Claude can interact with the GUI of any program that runs on your computer. In this instance, the API does not yet have functions for creating or labeling nodes. So Claude is using computer control to add the nodes and label them.
That's not me. That's Claude moving the mouse. Then it will use the API for the rest.
It looks like it has worked out which clips are ARRI, RED, and FILM based on their bins. While this isn't a feature of Resolve 21 .1, it's relevant to see how Claude can think outside the box to bridge gaps in Resolve's current MCP feature set. Yes, it would have been quicker to do these simple tasks myself.
But this is a really cool glimpse of what is possible.
AI is a complex and ever -evolving topic. It's hard to say anything definite without the privilege of hindsight. But here are some present thoughts.
Is AI ready to replace you? In short, no. We are very impressed with what is currently possible, and technology will continue to progress.
But AI creates derivative work, and that's not meant to be an insult. It finds patterns in the creativity of others and it replicates it. For now, creativity is and continues to be the domain of human beings.
In its present state, you can think of this sort of AI as an assistant editor. It requires a creative oversight, it needs well -considered instructions, and it needs someone to check its work. Thinking further, not all editing, though, is creative.
A lot of work is procedural. As AI editing capabilities progress, those procedural jobs will likely be the easiest to automate, but there is a definite limit to AI's creative ability. When a human being edits or is involved in any sort of creative performance, they draw on their emotional experiences and allow them to influence their decisions.
When we watch that content, we learn about and connect with the emotional experiences of the people who created it. A computer can imitate those emotional decisions, but it cannot make its own because it's not alive and it doesn't have its own experiences to draw on. Technological innovation affects jobs.
As an editor, you work in a competitive market where people value your speed, cost, or the quality of your work. If an AI agent can reliably complete a task quicker with greater quality or at a lower cost, then it might start to take work away from people. But that's not new.
If a fellow human editor can reliably complete a task quicker, with greater quality, or at a lower cost, then they will also take work away from other people. Whether a better AI tool or a better human editor comes along, it's possible to use scenarios like that as an opportunity to elevate and improve your own creativity and abilities.
Remember that true creativity is still a rare skill and one that is in demand.
AI is limited by the functionality that is currently exposed in DaVinci Resolve's API, and it's limited by the compute of your AI agent. For example, currently Blackmagic's API does not expose any of its IntelliSearch data to help your AI agent understand the visual content of your footage. And also, currently Cloud or ChatGPT does not have the ability, or more likely does not have the bandwidth, to watch your footage to understand its content.
We have no doubt that we will see Resolve's API functionality expand greatly and that in time we will see AI agents with greater capabilities. This is just the beginning. As always, we appreciate Blackmagic's approach.
AI has thus far been implemented in service of creativity rather than attempting to replace it. These demonstrations are a glimpse of what DaVinci Resolve is capable of and in part what Cloud is capable of. It's a smart division of labor.
It allows Blackmagic to focus on what they do best and it also allows you to choose the AI agent that performs best for your work.
If you're not sure what you can do, just ask your AI agent. It has access to Resolve's MCP and API documentation and can tell you what it can do or give you ideas to solve your problems. Complex prompts can take time for your AI agent to compute and Resolve will be inaccessible while your AI agent is processing those prompts.
it might be worth running those prompts on a second computer. For example, we have a Mac Mini M4. Its lack of power isn't an issue as most of the grunt work is done in the cloud.
If you only have one computer, run your prompts overnight and queue up your agent with as many requests as you can think of. This is just one of many new features in Resolve 21 .1. If you enjoyed this video, please give it a like, consider sharing it with a colleague, and make sure you're subscribed so you don't miss out on future content.
Thank you for watching.
The Hook
The bait, then the rug-pull.
DaVinci Resolve 21.1 just gave Claude and ChatGPT a direct line into its real editing API, not a screen-reader bolted onto the GUI. Team 2 Films spends the next nineteen minutes finding out exactly where that line ends.
Frameworks
Named ideas worth stealing.
13:24list
Four-node grading structure
Primary
LUT
Secondary
Look
The four serial color nodes the host asked Claude to create and label on every clip, matching each camera source (ARRI, RED, scanned film) to the correct LUT before grading.
Steal forany multi-camera grading session that needs a repeatable node order applied consistently across clips
CTA Breakdown
How they asked for the click.
VERBAL ASK
04:23product
“You can get an entire year of Audiio Pro for just $60 using our discount code in the description.”
Mid-roll sponsor read demoing the Audiio app UI, Elements 2.0 stems and the Hanz AI natural-language search, before cutting back to the Resolve demo.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
One editor runs nine real tasks through DaVinci Resolve's new native AI integration to find out where it saves real time and where it's just slower than doing it by hand.
How to connect ChatGPT to free DaVinci Resolve with an open-source MCP server so it can transcribe, cut, and caption a highlight reel, a feature Resolve normally locks behind its $300 Studio license.
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.
A content director runs 336 unsorted vlog clips through a Claude Code + DaVinci Resolve Studio pipeline that classifies A-roll from B-roll, proposes cutaway placements against four editorial rules, and drops the picks onto a real timeline — then shows exactly where it still needs a human.
00:21
00:41
00:55
01:17
01:34
01:46
01:54
02:30
02:43
02:58
03:12
03:25
03:44
03:59
04:03
04:30
04:46
04:59
05:14
05:34
05:48
06:03
06:17
06:31
06:45
07:04
07:14
07:25
07:42
08:11
08:19
08:39
08:51
09:08
09:22
09:50
10:05
10:19
10:33
10:47
11:07
11:12
11:30
11:58
12:15
12:27
12:41
12:50
13:10
13:24
13:38
13:52
14:21
14:35
14:49
15:04
15:19
15:48
16:02
16:17
16:29
16:43
17:14
17:27
17:40
18:09
18:23
18:51
17:33
24:14
13:44
17:02
15:15
16:54