Build your content engine with Claude Code and 14 tools, Quentin IA Marketing

A content idea can pass through five tools before you even start writing: research, transcription, editing, publishing, and tracking. The problem is not each individual tool. It is everything that happens between them.

In this system, Claude Code becomes the orchestrator. It reads your files, runs command-line tools, organizes the outputs, and stops when you need to make a decision.

You stop managing fourteen apps. You manage one folder, one instruction, and a few approval points.

The stack below covers the entire journey, from the initial angle to the finished video, landing page, and email. You do not need to install everything today. Start with the foundation, add one layer, test it end to end, then continue.

Claude AI Lab

The Claude AI Lab is my Skool community where I share my Claude systems and the more advanced modules. Access is $67/month.

Join the Lab →

✅What you need

Access to Claude Code. Claude Pro starts at $20 per month in the US and includes Claude Code. You can also use API billing.

A terminal and Node.js 18 or later. macOS, Linux, and Windows through WSL or Git Bash are supported.

A dedicated working folder. This is where your sources, instructions, raw files, and exports live.

Build the foundation

🧱 Claude Code + FFmpeg

Claude Code is the backbone. It is not a video editor. It is the agent that reads your project, executes commands, and moves files between the other tools.

Follow the official Claude Code setup guide, then launch it in your content folder:

npm install -g @anthropic-ai/claude-code
claude

These commands were verified locally with Claude Code 2.1.223 on August 6, 2026.

Next, add FFmpeg. It handles audio and video operations under the hood: cutting, converting, assembling, captioning, and rendering. On macOS with Homebrew:

brew install ffmpeg

Create a predictable structure:

content/
  research/
  raw/
  scripts/
  output/
  CLAUDE.md

The CLAUDE.md file loads automatically at startup. Only put information there that should remain true across sessions:

# Content context
- Audience: [YOUR_AUDIENCE]
- Main topic: [YOUR_TOPIC]
- Format: [YOUR_FORMAT]
- Voice: direct, concrete, jargon-free
- Approval: always show the script before rendering
- Outputs: save final files in output/

First test

Ask Claude to list the contents of your folder and summarize the rules in your CLAUDE.md. If it finds both without help, the foundation is ready.

Find the right angles

🔎 Apify + yt-dlp

Apify runs scrapers and exposes their results through an API. Its free plan currently includes $5 in credits, enough to test a lightweight research routine.

yt-dlp downloads media from many supported platforms. Install it with Homebrew or Python:

brew install yt-dlp
# or
python3 -m pip install -U "yt-dlp[default]"

📊

Use Apify to spot patterns

Collect search results, profiles, or comments with the right Actor, then save the raw data in research/.

📥

Use yt-dlp to collect

Download a video or its audio track when the platform and your usage rights allow it.

🧠

Use Claude to filter

Compare angles, hooks, and reactions. Ask it to cite the file and line that support each recommendation.

Run this first research pass:

Analyze the files in research/. Identify three angles that answer a concrete
question for [YOUR_AUDIENCE]. For each angle, provide the observed evidence,
a hook of no more than 12 words, and the recommended video format. Do not invent
missing data. Save the result to research/angles.md.

Guardrail

Work with public or authorized content and follow each platform's rules. A download tool does not turn protected or private content into material you are free to reuse.

Understand your footage

👂 Whisper + video-use

Whisper transcribes your audio and video files locally. Its code and model weights are released under the MIT License. FFmpeg is still required.

pip install -U openai-whisper
whisper raw/interview.mp4 --model turbo --output_dir research/

video-use goes further. It turns a conversation with an agent into editing decisions, removes filler, adds captions, and can generate overlays. Its code is open source, but the documented setup requires an ElevenLabs key for Scribe transcription. It is not a guaranteed zero-cost path.

📝

Start with Whisper

Choose it when your immediate need is local timestamped transcription with no API cost after setup.

✂️

Edit with video-use

Add it when you want to control cuts, captions, color, and overlays through a conversation.

🎯

Ask for a testable result

Request the strongest 15-second segment with its start time, end time, and rationale. You can verify the recommendation in seconds.

Recommended order

Start with Whisper. Add video-use only when raw/ already contains real footage and your output format is clear.

Make the video

🎬 4 visual engines

This layer has four options. They do not perform exactly the same job, and you do not need to adopt all of them.

HyperFrames turns HTML, CSS, and seekable animations into deterministic video. It is open source under the Apache 2.0 license. Remotion builds video with React and offers a free license for individuals and companies of up to three people. Editly assembles videos from a declarative configuration and FFmpeg. Higgsfield generates missing visual shots, with pricing that varies by region and usage.

🌐

HyperFrames

Choose it for animated cards, titles, product demos, and videos driven by HTML.

⚛️

Remotion

Choose it when your visual system already lives in React or you want to compose scenes like components.

🧩

Editly

Choose it to quickly sequence clips, images, titles, and music from a JSON file.

✨

Higgsfield

Keep it as a paid option for producing a cutaway shot you cannot film yourself.

Your first HyperFrames test:

Create a 6-second animated title card for [YOUR_TOPIC]. Use the colors and rules
from CLAUDE.md. Use a vertical 1080 by 1920 format. Show me a design screenshot
and wait for my approval before rendering the MP4 to output/.

Simple decision

Select one primary rendering engine. Keep the others as options for use cases it does not cover.

Prepare to publish

🚀 4 output tools

The last mile combines creation, manual review, and distribution.

Edits is Meta's mobile app for finishing and publishing. It exports without an added watermark and lets you check pacing, captions, and framing on a phone.

Claude Design creates visual prototypes, pages, and one-pagers through conversation. The product is currently in research preview for Claude Pro, Max, Team, and Enterprise subscribers.

Notion MCP lets Claude Code read and update your editorial calendar:

claude mcp add --transport http notion https://mcp.notion.com/mcp

Then run /mcp in Claude Code to complete OAuth authentication.

Resend handles transactional email through an API. Its free plan currently covers 3,000 emails per month, with a limit of 100 per day. Marketing campaigns have separate contact-based pricing.

📱

Edits validates the render

The final review stays human: mobile readability, cuts, music, safe zones, and overall feel.

🖼️

Claude Design builds the destination

Prepare the landing page or visual that turns attention into a next action.

🗓️

Notion keeps the record

Log the status, hook, files, scheduled date, and final URL in one entry.

✉️

Resend delivers the follow-up

Send the transactional email tied to a download, signup, or confirmation.

Connect the pipeline

🔗 1 master prompt

The minimum cost is simple: Claude Pro at $20 per month, then open-source tools and free plans. Higgsfield adds an optional expense that starts around $9 depending on the region. Apify, ElevenLabs, Remotion, and Resend can also become paid when your usage or team grows beyond their free limits.

💵

Minimum version

Claude Code, FFmpeg, yt-dlp, Whisper, and one open-source rendering engine. You can already research, transcribe, and produce.

🎥

Visual version

Add Higgsfield only when a generated shot genuinely improves the video.

⚙️

Automated version

Add Notion MCP and Resend when your pipeline runs often enough to justify automated distribution.

Install the system in this order:

🛠️Build plan

Foundation. Claude Code, FFmpeg, your folder, and your CLAUDE.md.

First loop. One authorized source, yt-dlp or Apify, then Whisper and an approved script.

First render. One visual engine, one MP4 export, and one phone review.

Distribution. Notion for tracking, then Resend or your email platform once the flow is stable.

Once every layer works on its own, give Claude Code this instruction:

You manage my content pipeline for [YOUR_AUDIENCE].

1. Analyze research/ and propose three angles supported by the available files.
2. Choose the most useful angle for [YOUR_GOAL] and write a [YOUR_DURATION]-second
   script that follows CLAUDE.md.
3. Stop and show me the hook, script, and scene plan.
4. Do not render anything without my explicit approval.
5. After approval, transcribe the footage in raw/, select the useful segments,
   create the visual elements, and export the video to output/.
6. Then prepare the landing page, transactional email, and Notion entry.
7. Finish with a report of the files created, costs incurred, and actions that
   still require my approval.

Do not invent sources, numbers, or publication statuses.

The rule that protects everything

Place a human checkpoint before the final script, before buying credits, and before publishing anything.

Your first piece of content should test the pipeline

Do not start with your most ambitious project. Use one short video, one angle, one piece of raw footage, and one title card. The result matters less than proving that every file reaches the right place and Claude stops when you asked it to.

Once that loop works, add one tool at a time. Your content engine becomes more powerful without becoming harder to operate.