Free guide ·
Connect Claude Code to Edits for Mac with its built-in MCP server, then direct videos in any niche, with Edits' own voiceover, captions and music.
By John Shemron, graphic designer and video editor at PromptVault
Instagram's Edits app for Mac has a built-in automation server. Turn it on, add it to Claude Code, and you can describe a video in plain words while Claude does the editing: it cuts the clips, writes on-screen text, adds Edits' own text-to-speech voiceover, generates captions, picks music and sound effects, checks the result and exports. You don't need to record a voiceover. You do need a Mac, Edits for Mac, a paid Claude plan, and something to show on screen, because Edits edits video but doesn't generate it.
This guide covers the five-minute setup, what Claude can and can't do inside Edits, one prompt you can reuse for any video, and recipes for six niches: cartoon stories, documentaries, film reviews, game reviews, faceless fact videos and product explainers.
/mcp.Edits says it plainly under the settings: anyone with the access token can edit your projects while the server is on. So never paste the token into a post, a screenshot or a chat. If it's ever exposed, click Regenerate and copy the Claude Code command again.
/mcp. The Edits server should be listed as connected. From Terminal, claude mcp list shows the same thing.Then test it with a question that changes nothing:
Which screen is Edits on right now, and what can you do from there?
Claude should answer with the app's current screen and the kinds of tools it can use. Edits only exposes its editing tools while a project is open, so if Claude can see just a few tools, open a project in Edits and ask again. If the list still doesn't update, start a new Claude Code session.
One rule before you start: Edits has one editor, and Claude edits whichever project is open in it. If two AI sessions drive Edits at the same time, or you switch projects by hand while Claude works, edits can land in the wrong project. In our own testing, a test project we had left open ended up holding another session's edit. Run one session at a time, and have Claude check the open project's title before it changes anything.
With a project open, Edits offered 285 actions on our Mac. They fall into these groups:
| Area | What Claude can do |
|---|---|
| Timeline | Import media, cut, split, trim, slip, reorder clips, layers and overlays, split screen, ripple edits |
| Text | Text layers with 34 presets, 77 animations and 416 fonts |
| Captions | Generate captions from speech, 39 caption styles, word highlights and emphasis |
| Voice | Text-to-speech with 27 built-in voices, or record a voiceover with a teleprompter |
| Music and sound | The Edits music library (For You, Trending, Original audio and Royalty-free), a sound effects library, volume, fades, ducking, voice enhancement, beat markers |
| Look | 77 filters, 16 LUTs, 148 video effects, colour grading, masks, 51 transitions |
| Motion | Keyframes with easing, crop, auto-reframe, object tracking, speed ramps, reverse |
| Checking | Capture frames, make a frame sheet of the whole video, describe a clip, detect shots, preview renders |
| Export | Export settings, start an export, check its progress |
The voices include calm, sage, poised, bold, friendly, cheerful and whisper, plus character voices such as robot, pirate, monster and chipmunk. On our Mac all 27 were US English.
What it can't do, as of 7 November 2026:
Because Edits doesn't generate pictures, every video starts with media you bring in. Four sources work well, and Claude Code can prepare most of them:
The clip above is the third kind, and it's what the Claude & Pixel Guide teaches: you paste one prompt into Claude Code and it builds a two-character dialogue Short with the same cast in every episode. It needs a Mac, and it's an independent course from PromptVault, not made by Anthropic or Meta. To see what else Claude can animate, read can Claude make animations?
Every recipe below follows the same order. Paste this into Claude Code and fill in the brackets:
Make a [length]-second vertical video in Edits about [topic], for [audience]. 1. Write the script as [number] short beats. Show me the script before you build anything. 2. Visuals: [where they come from: my clips in this folder / render them yourself with ffmpeg / import these AI clips]. Create a new Edits project and put them on the main track in order. 3. Voiceover: list the text-to-speech voices, pick a [calm / energetic / serious] one, and generate one voiceover per beat. Trim each clip to end about half a second after its line. 4. On-screen text: one short headline per beat in the upper third, with a bold preset and a short animation in. 5. Captions: generate them from the voiceover, choose a clean style with word highlights, and keep them in the lower third, clear of the headline. 6. Sound: a [mood] track from the Royalty-free music category, ducked under the voice and faded out at the end. Add a sound effect only where something changes. 7. Check: make a frame sheet of the whole video and look at it. Fix overlapping text, captions that run off screen, black frames and voice lines cut off mid-word. 8. Export it and tell me where the file is.
Two habits matter more than the exact wording:
The recipes below are starting points built on the tools above. We ran the documentary recipe end to end on our own Mac, and its notes list what it took. The other five haven't each been run end to end yet, so read the frame sheet before you export, and adjust.
Visuals from animation written as code, or from AI clips of a character you've kept consistent (how to keep the same character in AI animation covers that).
Write a 45-second cartoon story with two characters: a nervous robot and a calm cat. Five beats: the robot states a plan, the cat warns it, a flashback shows it failed before, the robot does it anyway, and the cat gets the last word. Render each beat as an animated scene in code, then build it in Edits: give each character its own text-to-speech voice (the robot voice for the robot, a calm one for the cat), captions with word highlights, a light comedy track from Royalty-free, and a short sound effect on the punchline.
Visuals from licence-clear photos online. Wikimedia Commons states the licence of every file, so Claude can keep only public domain, CC0 and CC BY photos and write the credits as it goes. We skipped share-alike licences.
We ran this one end to end on 7 November 2026: a 60-second documentary on the AI bubble, built entirely inside Edits by Claude Code.
The prompt, with the fixes from our run already in it:
Make a 60-second vertical documentary in Edits about the AI bubble. Keep it balanced: the case for a bubble and the case against. 1. Research first. Every number needs a source I can open. Show me the script, about 120 words in 8 beats, with the sources before you build anything. 2. Photos: find one licence-clear photo per beat on Wikimedia Commons (public domain, CC0 or CC BY only). Crop each to 9:16 and render it with ffmpeg as a ProRes clip with a slow zoom, as long as its voice line. Save the credits. 3. In Edits: create a new project and set it to SDR before you import anything. Clips in order, one text-to-speech line per beat, the stat as big text in the upper third with its source under it, captions in the lower third, short dissolves, a quiet music bed, and a photo credit line on the last beat. 4. Check: transcribe the finished voice and compare it with the script, fix any misheard captions, make a frame sheet, then export at 1080×1920.
The script it ran, 117 words, with the stat shown on screen for each beat:
| Beat | On screen | Source |
|---|---|---|
| The Nasdaq-100 fell 6.4% on 5 November 2026, its worst day since 2020 | −6.4% | CNN |
| About $2.5 trillion vanished; Nvidia alone lost about $450 billion | $2.5T | CNN |
| A year earlier, Nvidia became the first company worth $5 trillion | $5T | Associated Press |
| Big Tech plans to spend over $650 billion on AI in 2026 | $650B+ | Bloomberg |
| AI investment is about 3.6% of US GDP; the internet build-out was 1.1% | 3.6% vs 1.1% | The Wall Street Journal, via Wikipedia |
| Nvidia pledged up to $100 billion to OpenAI; OpenAI committed $1.4 trillion on about $13 billion of revenue | $1.4T vs $13B | Wikipedia: AI bubble |
| An MIT study found 95% of the companies it looked at got no measurable return from generative AI | 95% | MIT NANDA's 2025 report, via Campus Technology |
| Goldman Sachs and J.P. Morgan say the profits are real. Bubble or boom? | Bubble or boom? | Wikipedia: AI bubble |
What the run taught us, so your first try goes faster:
Every photo needs its credit. Ours, all from Wikimedia Commons and cropped: Nasdaq Tower May 2026 by Nielsoncaetanosalmeron (CC BY 4.0), Nvidia sign by Will Buckner (CC BY 2.0), Nvidia HQ by Kevin McCarthy (CC BY 2.0), Data centers in Ashburn by Theodore Christopher (CC0), Walking in Wall Street.jpg) by Ferfive (CC BY 4.0), Front of server racks at NERSC by Derrick Coetzee (CC0), Great Dome, MIT by Daderot (public domain), and New York Stock Exchange facade by Jeffrey Zeldman (CC BY 2.0).
For a documentary about a real person, keep to the same rules: use photos whose licence you can see, don't fake their face or voice, and check every fact before it goes in the script.
Visuals from score cards and text you control. Trailers and film clips belong to the studio, so if you use any, check the platform's rules and keep them short.
Make a 40-second review of [film]. Open with the verdict and a score out of 10 in the first two seconds. Then three beats: what works, what doesn't, who should watch it. Render a score card and three title cards with ffmpeg in a dark cinema style, use an energetic voice, captions with word highlights, a tense track from Royalty-free, and a whoosh on each beat change.
Visuals from your own gameplay recording. Many game publishers allow gameplay videos, but each sets its own rules, so check the game's.
Here's my gameplay recording: [file]. Find the five most action-packed moments (detect shots and describe the clips), and cut a 45-second review: verdict and score first, then the moments in order with one line of voiceover each. Add a speed ramp into the best moment, captions, a score card at the end, and an upbeat track from Royalty-free ducked under the voice.
Visuals from AI clips or backgrounds Claude renders. The facts are the product here, so give Claude the sources.
Make a 30-second "5 facts about [topic]" video. Use only facts from these sources: [links]. Show me the five facts with their sources before you build. One beat per fact, the number big on screen, a friendly voice, captions with word highlights, and a pop sound on each new number.
Visuals from a screen recording of your product.
Here's a screen recording of [product]: [file]. Make a 45-second explainer: the problem in one line, then three features. For each feature, zoom in on the part of the screen that matters with keyframes, add a short callout, and explain it in one line of voiceover. End on the product name and one call to action. Use a calm voice and a light track from Royalty-free.
Edits' text-to-speech replaces recording a voiceover:
All 27 voices on our Mac were US English. For another language or a recurring character with a voice of its own, generate the audio elsewhere and import it; free AI voices and lip sync for cartoon characters covers free voice models.
No. The automation server is in Edits for Mac. Export the finished video there and post it as you normally would.
No. Edits has text-to-speech built in, with 27 voices on our Mac, and Claude can generate one voiceover per line of your script and caption it.
No. The Edits automation server has no posting or scheduling tools. Claude exports the video, and you post it.
No. Edits edits and finishes video. The pictures come from your own clips, AI video tools, or animation and graphics Claude Code renders with code.
Edits' automation settings also offer setup commands for Codex and Muse. The prompts in this guide aren't specific to Claude, but we've only connected Claude Code.
For the bigger picture of making videos with Claude Code, see how to make videos with Claude Code and the prompt animation guide.
The Claude & Pixel Guide: Animated dialogue videos for Reels, Shorts and TikTok, made in code from one prompt. 15 text lessons in three tracks.
