The expensive part of content was never the software. It was the twelve context switches between opening the software.
Here is the direct answer. Two paid subscriptions, Claude Pro at $20 and Masonry Starter at $10, plus eleven tools that are open source or free at the volume one person produces. Claude Code is not another content tool in that list. It is the layer that turns the other twelve from a folder of bookmarks into a pipeline that runs end to end from a single brief.
The Whole Stack, In One Table
| Tool | Job in the pipeline | Cost |
|---|---|---|
| Claude Code | Connects everything, writes the glue | $20 |
| Masonry | Generates B-roll, images, footage | $10 |
| Remotion | Claude can animate, in React | Free* |
| Hyperframes | HTML to MP4, agent-native | Free |
| video-use | Claude can read a video timeline | Free* |
| Whisper | Claude can hear | Free |
| Editly | Declarative command line editing | Free |
| Edits | Captions and quick mobile trims | Free |
| Apify | Researches transcripts at scale | Free* |
| yt-dlp | Pulls source video and subtitles | Free |
| Claude design skills | Landing pages for each piece | Free |
| Notion | Content calendar and state | Free |
| Resend | Email and lead list | Free* |
| Total, solo operator | $30/mo | |
* Four entries have a threshold worth knowing before you build on them. They are covered in Where the Free Tier Runs Out rather than buried in a footnote.
The Layer That Makes It a System
Orchestration
Every other tool here is a command line program, a library, or an API. That is not incidental, it is the whole design. Agents are poor at driving graphical timelines and excellent at writing and running code, so a stack built entirely from things with a CLI is a stack an agent can actually operate.
Claude Code reads your brief, calls yt-dlp for source footage, runs Whisper for a transcript, writes the Remotion or Hyperframes composition, triggers the render, and files the output in Notion. Included in Claude Pro at $20. It does have usage limits, and if you hit them mid-render the fix is workflow rather than money: I wrote up the commands that cut the token burn separately.
Generation: Making Something Out of Nothing
B-roll, images, footage
Starter is $10 a month for 1,000 credits, which works out to roughly 500 images or about 33 videos, with access to all models and three generations running at once. There is a free tier of 50 credits if you want to test the shape of it first. For a solo publisher shipping a few pieces a week, the credit budget is the binding constraint, not the model quality.
Landing pages
Every serious piece of content eventually needs a page to point at. Doing this inside the same agent that made the video means the page and the video share a brief rather than drifting apart. Covered in more depth in the design tools post.
Input: How Claude Sees and Hears
Source acquisition
The workhorse. Open source, actively maintained, pulls video, audio, and subtitle tracks from a very long list of sites. This is the front door of the pipeline: nothing downstream runs until something has been fetched. Obvious but necessary caveat, downloading is governed by the source platform's terms and by copyright, and "my agent did it" is not a defence.
Transcription
OpenAI's open source speech recognition model. Free in the sense that matters, no API bill, but you pay in compute: it runs on your machine, and long files on a laptop are slow. Use it for bulk transcription where latency does not matter and quality is good enough.
Timeline comprehension
An open source Claude Code skill from the browser-use team. You drop raw footage in a folder, talk to Claude Code, and get final.mp4 back: clips cut, filler words removed, subtitles added, colour graded, rendered.
The interesting detail, and the one that corrects a common misreading: the model never watches the video. It reads it. One ElevenLabs Scribe call per source returns word-level timestamps, speaker diarization, and audio events, and that transcript layer is what lets it cut on word boundaries rather than guessing at frames. A word-accurate timeline is more useful for editing than a pixel-accurate one, which is why this design works. It also means the skill needs an ELEVENLABS_API_KEY, and that is the hidden cost in this stack.
Research at scale
Where yt-dlp fetches one thing well, Apify sweeps many. Useful for pulling transcripts across a whole channel or competitor set so Claude has a corpus to reason over instead of a single video. The free plan gives $5 of prepaid platform usage a month with no card required, up to 16GB of actor RAM, and 25 concurrent runs. Unused credit does not roll over.
Output: Four Ways to Turn Code Into Video
This is where the stack looks redundant and is not. Four rendering tools, four genuinely different jobs.
Remotion
React based. A composition is a component, animation is state as a function of the frame number. Best when your video has data-driven variants and you already write React.
Hyperframes
Plain HTML, CSS, and data-* attributes, captured frame by frame and encoded with FFmpeg. Deterministic: the same input gives the same output every time.
Programmatic animation
Make videos in React, then render them from the command line. The licence is the thing to read before you commit: free for individuals, non-profits, and for-profit teams of up to three people, commercial use included. Past that you need a company licence, and there is no feature difference between them, only a headcount threshold.
HTML to MP4
Open sourced by HeyGen and explicitly built to be agent-native, which is the correct framing. Coding agents are hopeless at navigating Premiere Pro and fluent in HTML and JavaScript, so defining a video as a web page removes the impedance mismatch. Clips, subtitles, images, audio, and timing all live in HTML elements and data attributes. Animate with CSS, GSAP, Lottie, or Three.js, preview in a browser, then render to MP4: the renderer seeks each frame in headless Chrome and encodes the result with FFmpeg. The CLI is non-interactive by default, which is the detail that makes it scriptable by an agent rather than a person.
Declarative assembly
A Node.js and FFmpeg framework for declarative non-linear editing. You hand it a JSON description of clips, images, titles, transitions, and audio, and it assembles the video. It streams rather than writing intermediate files, so it is fast and light on disk. This is the right tool when the edit is structural rather than creative: montages, batch variants, templated openers.
Captions and the human pass
Instagram's free editing app, shipped in April 2026 and updated at an unusual clip since. Auto-generated captions with real styling control, plus the quick trims you want to make with your thumbs rather than by writing a config file. Every automated pipeline needs one manual station at the end, and on mobile this is a good one.
Distribution: The Half Everyone Skips
Calendar and state
Less glamorous than the render pipeline and more load-bearing. A pipeline needs somewhere to record what exists, what is scheduled, and what shipped. Free tier is comfortably enough for one person, and Claude can read and write it directly, which is the part that matters.
Email and lead list
Free plan is 3,000 emails a month, capped at 100 a day, on one domain. Every tier including free gets the same core feature set: REST API, SMTP relay, SDKs, inbound handling, batch sending, open and click tracking, React Email, and DKIM, SPF, and DMARC authentication. That daily cap is the one to plan around.
Where the Free Tier Runs Out
Three thresholds, in the order you will hit them. None of these are dealbreakers. All of them are cheaper to know now than to discover mid-launch.
- 01Apify, at $5 of credit. Prepaid, monthly, does not roll over. Fine for weekly research. Not fine for a one-off sweep of two hundred channels, which will exhaust it in an afternoon.
- 02Resend, at 100 emails a day. The monthly 3,000 sounds generous until you try to send a single broadcast to 400 subscribers and it takes four days. Plan sends, or budget $20 for Pro.
- 03ElevenLabs, at the commercial licence. The free tier is 10,000 credits a month with no commercial licence, so the moment your output is monetised you need Starter at $6. Scribe via the API runs $0.22 an hour for standard transcription, and is materially cheaper than the same job through the web interface.
- 04Remotion, at your fourth teammate. Free for individuals, non-profits, and teams up to three. Past that: Creators at $25 per seat per month, or Automators at $0.01 per render with a $100 monthly minimum if you are building an automated pipeline, which this stack arguably is.
So what does it really cost? For one person doing commercial work: $20 Claude Pro + $10 Masonry + $6 ElevenLabs Starter = $36 a month. The $30 headline is accurate for the two subscriptions and honest about everything except the transcription licence. For a team of four or more, add a Remotion company licence and you are into three figures, which is still less than one freelance editor invoice.
How the Pieces Actually Chain
# 1. acquire yt-dlp <source-url> --write-auto-subs # 2. understand whisper input.mp4 --model medium --output_format json # 3. decide (Claude Code reads the transcript, # picks the beats, writes the composition) # 4. generate the missing footage masonry → b-roll for beats 2, 5, 9 # 5. render npx remotion render # data-driven variants npx hyperframes build # html compositions # 6. ship notion → mark scheduled resend → notify the list
Step 3 is the only one that needed a human two years ago and is the only one worth your attention now. Everything above and below it is plumbing, and plumbing is precisely what an agent should own.
The stack is not impressive because it is cheap. It is impressive because every piece has a command line, which is the only reason an agent can run any of it.
What I Would Actually Do First
- 01Do not build all thirteen. Build yt-dlp to Whisper to Claude Code and stop. If the transcript-to-outline step does not save you real time, nothing downstream will.
- 02Pick one renderer. Remotion if you write React, Hyperframes if you would rather your agent wrote HTML. Running both from day one is how this becomes a hobby instead of a system.
- 03Keep one manual station. Fully automated content reads like fully automated content. The captions pass is a good place to put a human, because it is where taste is cheapest to apply.
- 04Wire Notion early, not last. A pipeline with no state produces things nobody can find, which is indistinguishable from producing nothing.
The broader point is that the cost of production has collapsed and the cost of judgement has not moved. Thirty dollars a month buys you the ability to make ten times as much content, which is only an advantage if you had something worth saying at volume one. The stack is real, it works, and it will happily help you publish nothing of value very efficiently.
Questions People Ask
Is this content stack really only $30 a month?
Two line items are paid: Claude Pro at $20 a month, which includes Claude Code, and Masonry Starter at $10 a month for image and video generation. That is a genuine $30. The other eleven tools are open source or have free tiers that a solo operator will not exhaust. The honest total is closer to $36 once you add ElevenLabs Starter at $6, which you need for commercial rights on the transcription that video-use depends on.
What does Claude Code actually do in this stack?
It is the orchestration layer, not another content tool. Every other item here is a command line program, a library, or an API. Claude Code is the thing that reads your brief, calls yt-dlp to pull source video, runs Whisper for the transcript, writes the Remotion or Hyperframes composition, invokes the render, and files the result in Notion. Without it you have thirteen tools. With it you have one pipeline.
Is Remotion free for commercial use?
It is free for individuals, non-profits, and for-profit teams of up to three people, including commercial work. Past that you need a company licence: Remotion for Creators at $25 per seat per month, or Remotion for Automators at $0.01 per render with a $100 monthly minimum if you are building an automated video pipeline. There is no feature difference between the free and paid versions, only a headcount and use-case threshold.
What is the difference between Remotion and Hyperframes?
Both turn code into video, but from different starting points. Remotion is React based, so a composition is a component and animation is state over frames. Hyperframes is plain HTML, CSS, and data attributes, rendered frame by frame and encoded with FFmpeg, and it is explicitly built to be agent-native. If your agent is more fluent in HTML than React, or you want deterministic identical output on every render, Hyperframes is the lower-friction choice.
Does Claude actually watch the video?
Not in the way most people assume. In video-use the model never watches raw frames. It reads the video through a transcript layer: one ElevenLabs Scribe call per source returns word-level timestamps, speaker diarization, and audio events, which is what lets it cut on word boundaries. That is a design choice rather than a limitation, because a word-accurate timeline is more useful for editing than a pixel-accurate one.
Where does the free tier run out first?
Three places, in this order. Apify gives you $5 of prepaid platform credit a month and it does not roll over, so heavy transcript scraping stops early. Resend's free plan is 3,000 emails a month but capped at 100 a day on one domain, which breaks the first time you send a real broadcast. And ElevenLabs' free tier carries no commercial licence, so the moment your output is monetised you need at least the $6 Starter plan.
Want this built rather than bookmarked?
I help founders and brands build generative AI systems that ship: model routing, agentic workflows, coding-agent setups, and content engines that are production-grade down to the plumbing. If you want a second pair of eyes on yours, book a session.
Book a Growth Chat
Prices checked July 27, 2026 and they move, so verify before you commit a budget. Masonry tier pricing and credit allowances are from masonry.so/pricing. Remotion's licence terms, the three-person threshold, and both company licence structures are from Remotion's own licence documentation. Hyperframes' architecture, default output settings, and agent-native design intent are from the HeyGen open source project and its documentation. video-use's operation, including the ElevenLabs Scribe dependency and the fact that the model reads a transcript layer rather than watching frames, is from the browser-use repository. Editly's capabilities are from the mifi/editly repository. Apify free-plan credit, RAM, and concurrency limits, and Resend's free-tier volume, daily cap, and feature parity across tiers, are from those companies' pricing pages and documentation. ElevenLabs free-tier credits, the absence of a commercial licence on the free plan, Starter pricing, and Scribe API rates are from ElevenLabs pricing coverage. Two disclosures and one ambiguity. Masonry is my own company, so treat that line as an interested recommendation rather than a neutral one; it replaced a comparably priced tool in the same slot and the $30 total is unchanged either way. Nothing else here is sponsored. The ambiguity: the source list for this post read "16 Edits app" for the captions tool, which I have taken to mean Instagram's Edits app, since that is free and matches the described job. If a different tool was meant, that one line is the one to correct.