How to Build an AI Content Engine with Claude Code

The System That Changed Everything

I went from spending 4 hours per video to mass producing 23 scripted videos in a single sitting.

Not hypothetical. Not "potential output." I sat down one morning, opened my terminal, and by lunch I had 23 videos with researched hooks, scripted talking points, and ready-to-film cards sitting in my pipeline. All I had to do was press record.

The system behind it is something I call a content engine. It lives inside Claude Code, which is an AI that runs in your terminal, reads your files, and executes real commands on your computer. It's not a chatbot. It's closer to having a research assistant, content strategist, and scriptwriter sitting inside your laptop, waiting for instructions.

The engine handles five stages of content creation, all from one terminal window. Research, multiply, script, film, post. Each stage feeds the next. You control the creative decisions. The AI does the tedious work that used to eat your entire day.

This is the workflow:

Every step in that diagram is a single command you type into your terminal. No switching between 8 tabs. No copy-pasting between tools. One system, one place, every day.

This guide walks you through building the entire thing from scratch. I'm assuming you've never opened a terminal before. By the end, you'll have a working AI content engine that researches trends, generates video ideas, writes scripts in your voice, and can publish across every platform you use.

I'm giving away the full strategy here. How it works, why each piece matters, and exactly how to set it up. The only thing behind the email wall is the template files that save you a few hours of configuration.

Let's build it.

What You Need Before You Start

You don't need to know how to code. You don't need a computer science background. If you can type and follow instructions, you can build this. But you do need a few things ready before we start.

Hardware

A computer. Mac, Windows, or Linux all work. Claude Code runs in the terminal, so if your machine can open a command line, you're good. No special specs required.

Claude Code Account

Claude Code is free to install. To actually use it, you need either a Claude Pro subscription ($20/month through Anthropic) or a pay-as-you-go API key where you're billed based on usage. The Pro subscription is the simplest path if you're just getting started. You'll set this up in Step 1.

Free Tool Accounts

The content engine connects to three external tools. All of them have free tiers that are more than enough to start:

• Airtable (free tier) - This is your content pipeline. Every idea, script, and published post lives here in one organized view. Think of it as a spreadsheet that actually makes sense.
• Supabase (free tier) - This is your analytics database. When you start analyzing what content performs and why, the data goes here. You won't need this on day one, but setting it up now takes 5 minutes and saves you time later.
• Apify (free tier with monthly credits) - This is how the engine scrapes the internet for research. It pulls trending topics from X, Reddit, GitHub, and news sites so you always know what's worth making content about.
• Dropbox (free tier) - This is where your edited videos live. Drop a video into a folder, and the posting skill pulls it automatically. You'll create a developer app to connect it.
• Late (free tier) - This is how the engine publishes to every platform at once. Late connects your Instagram, TikTok, YouTube, LinkedIn, X, and Threads accounts into one API. One call, all platforms.

Optional (But Useful)

• yt-dlp - A free command-line tool that downloads videos from any platform. Useful when you want to analyze competitor content or pull reference clips.
• ffmpeg - A free tool for processing video and audio files. The posting skill uses this to extract audio for transcription.
• Whisper - OpenAI's free transcription model. Runs locally on your machine. The posting skill uses it to transcribe your video before generating captions.

Time

About 30 minutes for initial setup if you follow this guide straight through. After that, the daily workflow takes 15-20 minutes. Most of that time is you making creative decisions, not doing busywork.

Step 1: Install Claude Code

Before we touch any content strategy, we need the brain of the operation running on your machine.

What Claude Code Actually Is

You've probably used ChatGPT or Claude in a browser. You type a message, it responds, and that's the end of it. Claude Code is different. It's an AI that lives in your terminal (the command-line interface on your computer), can read and write files on your machine, run commands, and connect to external tools through something called MCP servers. When you tell it to research a topic, it actually goes and scrapes real websites. When you tell it to write a script, it saves a real file to your computer. When you tell it to update your content pipeline, it actually modifies the database. It does things. That distinction matters.

Create Your Anthropic Account

1. Go to claude.ai and create an account if you don't have one
2. Subscribe to Claude Pro ($20/month) or set up API billing at console.anthropic.com

Claude Pro is the easiest route. You get a generous amount of usage per month and don't need to worry about per-message billing. If you prefer pay-as-you-go, the API works too, but you'll need to add a credit card and manage your own spending limits.

Open Your Terminal

If you've never done this before, don't overthink it. The terminal is just a text-based way to talk to your computer.

On Mac: Press Cmd + Space, type "Terminal," and hit Enter. A window with a blinking cursor will appear. That's it.

On Windows: Press the Windows key, type "Command Prompt" or "PowerShell," and open it. Same idea.

On Linux: You already know where your terminal is.

Install Claude Code

Type this command into your terminal and press Enter:

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

If you get an error saying npm is not found, you need to install Node.js first. Go to nodejs.org, download the LTS version, run the installer, then try the command again.

Once it finishes installing (might take a minute), verify it works:

claude

Claude Code should start up and greet you. It will ask you to authenticate with your Anthropic account the first time. Follow the prompts in your terminal to log in.

If you see a prompt waiting for your input, you're live. Type /exit to close it for now. We need to set up your project before we start using it for real.

Important: This is NOT the Claude chatbot you use in a browser. That's Claude the conversational AI. This is Claude Code, the command-line tool that can actually interact with your computer and external services. Same company, different product. Don't mix them up.

Step 2: Set Up Your Project

Claude Code reads files from whatever folder you run it in. So the first thing we need is a dedicated folder for your content system, with one specific file that tells Claude who you are, what you do, and how you work.

Create Your Content Folder

In your terminal, run these commands one at a time:

mkdir -p ~/content-engine
cd ~/content-engine

That creates a folder called content-engine in your home directory and moves you into it. This is where everything will live: your config, your scripts, your reference files, your pipeline data.

Create Your CLAUDE.md File

This is the single most important file in your entire system. CLAUDE.md is a configuration file that Claude Code reads every time it starts a session. It tells the AI who you are, what platforms you post on, what your voice sounds like, where your tools are connected, and what rules to follow.

Think of it as a briefing document. Every morning when you open Claude Code, it reads this file first and loads your entire context into memory. Without it, Claude is a general-purpose AI. With it, Claude is YOUR content strategist who knows your brand, your audience, and your workflow.

Create the file:

touch CLAUDE.md

Now open it in any text editor (TextEdit on Mac, Notepad on Windows, or VS Code if you have it) and fill in the template. The structure looks like this:

# My Content Engine

## Business Config
- Business name: [Your name or brand]
- What you do: [One sentence about your content/business]
- Platforms: [List every platform you post on with handles]

## Brand Voice
- Tone: [3-4 adjectives that describe how you sound]
- Always: [What your content should feel like]
- Never: [What your content should NOT feel like]

## Niche Keywords
[List 10-15 keywords/topics you cover]

## Tools
- Airtable Base ID: [filled in Step 3]
- Supabase Project: [filled in Step 3]
- Apify Token: [filled in Step 3]

Be specific in the voice section. "Professional but friendly" is useless. "Talks like a 26-year-old who reads a lot and swears occasionally, mixes internet slang with business concepts, never sounds like a LinkedIn influencer" is useful. The more specific you are, the better Claude writes for you.

The template pack (available at the bottom of this guide) includes a fully filled-out CLAUDE.md with every section you need. But even this bare-bones version will get you started.

Step 3: Connect Your Tools

The content engine needs three external services to work at full power. These are the hands and eyes of the system. Claude Code is the brain that coordinates everything, but it needs Airtable to organize, Supabase to analyze, and Apify to research.

Each tool connects through something called an MCP server (Model Context Protocol). You don't need to understand how MCP works under the hood. Just know that when you run the connection commands below, you're giving Claude Code permission to read and write data in these tools directly from your terminal.

Airtable (Your Content Pipeline)

Airtable is where every piece of content lives, from raw idea to published post. The engine creates records automatically as you move through the workflow. An idea gets logged, then promoted to a script, then marked as filmed, then published. One board, every stage visible at a glance.

1. Go to airtable.com and create a free account
2. Create a new base (Airtable's word for a project/database)
3. If you have the template pack, import the CSV file I provide. It pre-builds all the columns and views you need. If not, create a table with these columns: Title, Status (Idea / Scripted / Filmed / Published), Platform, Hook, Script Notes, Date Created, Date Published.
4. Go to airtable.com/create/tokens and generate a personal access token with read/write permissions on your base

Now connect it to Claude Code. In your terminal, inside your content-engine folder, run:

claude mcp add airtable -- npx -y @anthropic/claude-airtable-mcp --api-key YOUR_AIRTABLE_TOKEN

Replace YOUR_AIRTABLE_TOKEN with the actual token you just generated. Once this runs, Claude Code can read your pipeline, create new entries, and update statuses. All from the terminal.

Supabase (Your Analytics Database)

Supabase stores your content performance data. When you start analyzing which hooks get the most views, which formats drive engagement, and what posting times work best, this is where that data lives. It powers the analysis skills that help you make better content over time.

1. Go to supabase.com and create a free account
2. Create a new project (pick any region close to you, set a database password you'll remember)
3. Once the project is ready, go to the SQL Editor in the Supabase dashboard
4. Run the migration SQL from the template pack (it creates the tables for storing content metrics, hook performance, and engagement data). If you don't have the template pack yet, skip this for now. You can add it later.

The MCP connection for Supabase is pre-configured when you install the Claude Code Supabase integration. Follow the setup prompts when you first run a skill that needs analytics data.

Apify (Your Research Scraper)

Apify is the research arm. When the content engine scrapes X for trending AI discussions, pulls Reddit threads about your niche, or checks GitHub Trending for new tools, Apify is doing the actual web scraping behind the scenes.

1. Go to apify.com and create a free account
2. The free tier gives you monthly credits that are more than enough for daily research scraping
3. Go to Settings > Integrations and copy your API token

Add your Apify token to your environment variables so Claude Code can access it. On Mac/Linux, add this line to your ~/.zshrc or ~/.bashrc file:

export APIFY_TOKEN="your_token_here"

Then restart your terminal or run source ~/.zshrc to load it.

Update your CLAUDE.md with the Airtable base ID and Supabase project URL so the engine knows where to point. The exact fields are in the template.

Step 4: Install the Skills (The Content Engine)

This is where the system comes to life.

Skills are reusable instruction sets that Claude Code follows when you call them. Think of each skill as a detailed job description for a specific task. When you type /daily-content-researcher, Claude doesn't wing it. It reads the skill file, follows the exact research process you defined, scrapes the sources you specified, applies the ranking criteria you set, and outputs the results in the format you designed.

Without skills, Claude Code is smart but undirected. With skills, it runs your content operation the same way every single time.

The content engine uses six skills. Each one handles a different stage of the workflow.

1. /daily-content-researcher

Scrapes four sources every morning (X, Reddit, GitHub Trending, and tech news), filters for topics in your niche, and presents a ranked Top 10 list of what's worth making content about today. Each topic comes with a brief on why it's trending, how large the audience is, and a suggested angle. You pick your favorites, and they auto-save to your Airtable pipeline as Ideas.

2. /content-ideator

Takes a single topic from your pipeline and goes deep. Researches it across multiple sources, looks at what content already exists about it, identifies the angles nobody has covered yet, and generates five distinct video variations. Each variation has a different hook angle, a different audience entry point, and a different emotional driver. One topic becomes five videos, and every one of them has a reason to exist.

3. /content-scripter

The workhorse. For each video, it pulls hook performance data from your analytics, presents six data-backed hook options ranked by estimated performance, and lets you pick. Then it generates a filming card: a structured outline with your hook, three talking points, a close, and a CTA. Not a word-for-word script (those sound robotic on camera), but a tight structure you can riff from naturally.

4. /shortform-analysis

Analyzes your own published content to find patterns. Which hooks get the most views? Which video lengths drive the best retention? What posting times work for your audience? It pulls your performance data from Supabase, runs the analysis, and gives you plain-English recommendations. Run this weekly and your content gets smarter over time.

5. /competitor-analyst

Point it at any creator or brand, and it breaks down their content strategy. Posting frequency, hook patterns, format mix, engagement trends, top-performing content and why it worked. The goal isn't to copy anyone. It's to understand what structures and patterns are working in your space so you can apply them with your own voice and angle.

6. /video-downloader

Downloads videos from any platform (YouTube, TikTok, Instagram, X) directly to your computer. Useful for competitor analysis, building reference libraries, or pulling clips you want to study. One command, any URL.

How to Install Skills

Skills live in a hidden folder on your computer called ~/.claude/skills/. Each skill is a markdown file with specific instructions Claude Code reads and follows. You create the folder structure, drop the skill files in, and Claude Code picks them up automatically.

mkdir -p ~/.claude/skills/daily-content-researcher
mkdir -p ~/.claude/skills/content-ideator
mkdir -p ~/.claude/skills/content-scripter
mkdir -p ~/.claude/skills/shortform-analysis
mkdir -p ~/.claude/skills/competitor-analyst
mkdir -p ~/.claude/skills/video-downloader

Each folder gets a SKILL.md file that contains the full instructions for that skill. The template pack includes all six, pre-written and ready to drop in.

You can also write your own skills for anything. Client-specific workflows, email sequences, blog post outlines, podcast prep. The format is the same: a markdown file that tells Claude Code exactly what to do, step by step.

Step 5: Set Up Your Reference Files

Skills tell Claude what to DO. Reference files tell Claude who you ARE.

These files live in your project folder and get loaded whenever Claude writes content for you. They contain your voice, your audience profile, your proven content patterns, and your hook templates. Fill them in once. Claude reads them every time it writes, so your scripts always sound like you wrote them.

avatar.md

This file describes your audience in two tiers. Your MACRO audience is the broad group of people who consume your content casually: they're scrolling, they want something interesting, and they might follow you if you consistently deliver. Your MICRO audience is smaller and more specific: these are people who might actually buy something from you, hire you, or become a client.

Describe both groups. Age range, where they hang out online, what they care about, what pain they have, what kind of content makes them stop scrolling. The more specific you get, the better Claude targets your scripts.

scripting-voice.md

This is where you teach Claude how you talk on camera. Not how you write, but how you speak. Include your catchphrases, the way you open videos, words you use all the time, words you never use, your energy level, and any verbal tics that make you sound like you.

Record yourself talking for 5 minutes about anything. Transcribe it (Claude Code can do this with Whisper). Read the transcript and note what makes your speech patterns yours. Put that in this file. The result is scripts that sound like you talking, not like an AI writing for a generic creator.

viral-content-patterns.md

There are roughly eight content archetypes that consistently go viral in the short-form video space. Things like "impossible demo" (showing AI doing something viewers didn't think was possible), "versus comparison" (tool A vs. tool B), "speed challenge" (building something in a time limit), and others. This file catalogs the archetypes with examples and notes on when each one works best.

You build this over time by paying attention to what blows up in your niche. Every time you see a video with outsized engagement, note the pattern. After a few weeks, you'll have a reference library that helps Claude generate ideas in formats that are proven to perform.

hook-swipe-file.md

A collection of hook templates organized by archetype. "I tested [tool] for 30 days and the results are insane." "Nobody is talking about this AI tool that does [thing]." "Stop using [popular tool] for [task]. Do this instead."

These aren't scripts. They're proven structures the /content-scripter skill uses when generating hook options for your videos. The more hooks you add to this file (from your own top performers and from competitors), the better the hook suggestions get.

Create these files in your project folder:

touch ~/content-engine/reference/avatar.md
touch ~/content-engine/reference/scripting-voice.md
touch ~/content-engine/reference/viral-content-patterns.md
touch ~/content-engine/reference/hook-swipe-file.md

(You'll need to create the reference folder first: mkdir -p ~/content-engine/reference)

You don't need all four files to be perfect on day one. Start with avatar.md and scripting-voice.md. Those two alone will make a noticeable difference in how well Claude writes for you. Build the other two over time as you collect data.

Step 6: Run Your First Session (Full Walkthrough)

Everything is installed. Tools are connected. Skills are loaded. Reference files are in place. Time to run the engine.

Open your terminal, navigate to your project folder, and start Claude Code:

cd ~/content-engine
claude

Claude loads your CLAUDE.md, recognizes your skills, and waits for instructions. From here, the daily workflow follows five stages. I'll walk you through each one as if this is your first time.

Stage 1: Research

Type:

/daily-content-researcher

Watch your terminal come alive. Claude reaches out to X, Reddit, GitHub Trending, and tech news sources through Apify. It pulls back dozens of trending topics, filters for your niche keywords, and ranks them by three criteria: total addressable audience (how many people care about this topic), demo-ability (can you show it on camera?), and hook potential (is there a surprising angle?).

After a minute or two of scraping and processing, it presents a Top 10 list. Each entry has a title, a brief on why it's trending, the source it came from, and a suggested content angle.

You read through the list. Maybe today's top hits are a new Claude Code feature release, a viral AI demo on X, and a Reddit thread about AI replacing content agencies. You pick your 3-4 favorites. Claude saves them to your Airtable pipeline as Ideas with status "Idea" and today's date.

Stage 2: Multiply

Type:

/content-ideator

Claude pulls your Ideas from Airtable and asks which ones you want to develop. Pick one (or all of them). For each topic, it does a deeper research pass: reads the source material, checks what content already exists about the topic, and identifies angles that are missing or underserved.

Then it generates five video variations per topic. Not five copies of the same video. Five genuinely different approaches. One might be a "speed build" demo. Another might be a "reaction to bad takes" format. A third might be a tutorial angle. Each variation has a distinct hook angle, a target audience (MACRO or MICRO), and a one-line thesis statement.

If you started with 4 topics, you now have 20 potential videos. Pick the ones that excite you. Claude updates each one in Airtable with its details and moves them to "Ideated" status.

Stage 3: Script

Type:

/content-scripter

This is where the magic happens. Claude pulls your Ideated videos and, for each one, does three things.

First, it checks your hook-swipe-file.md and your performance data (if you have analytics in Supabase) to find hook structures that work for this type of content. It generates six hook options and presents them to you ranked by estimated performance.

You pick the hook that hits hardest.

Then Claude generates a filming card. Not a full script that you'd read word-for-word (that sounds dead on camera). A structured outline with your hook, three talking points (each with a single data point or example to reference), a close, and a CTA. The talking points are designed as springboards. You read the card, internalize the structure, and riff naturally on camera.

Each filming card also includes framing notes: where to look, whether to screen-record a demo, and a suggested video length based on the format.

If you scripted all 20 variations, you now have 20 ready-to-film video cards. Each one researched, hooked, and structured. All sitting in your Airtable pipeline. Status: "Scripted."

Stage 4: Film

This part is you. Press record, reference your filming card, and talk. The engine doesn't film for you (yet). But the prep work that used to take hours is already done. You know what you're saying, why you're saying it, and how to open strong.

Most people can batch-film 5-10 videos in a single session once the scripts are ready. That's a week or two of daily content, shot in one afternoon.

Stage 5: Post

Type:

/post-content

This is the part that feels like cheating. You drop your edited video into a Dropbox folder called /Content/ready/. That's it. That's your upload.

Claude scans the folder, shows you what's queued with preview links, and asks which video you want to post. Pick one. It downloads the video, runs it through Whisper for a transcript, then generates platform-specific captions using your voice profile. Instagram gets a one-liner with hashtags. LinkedIn gets a 2-3 line professional angle. Twitter gets the same energy without hashtags. Every caption matches how you actually write.

You review the captions. Tweak if you want. Say "post" and Claude publishes to every connected platform in a single API call. Instagram Reels, TikTok, YouTube Shorts, LinkedIn, X, Threads. One command, seven platforms, posted simultaneously.

After publishing, it moves the video from /ready/ to /posted/ so your queue stays clean. You get back a list of live links to every published post.

The whole thing — from "I have an edited video" to "it's live on seven platforms" — takes about 60 seconds. No logging into each app. No copy-pasting captions. No uploading the same video file seven separate times.

That's the loop. Research, multiply, script, film, post. Every day. The AI does the 80% that's tedious. You do the 20% that requires your face, your voice, and your personality.

What This Looks Like in Practice

Numbers help, so here's what a real session produced.

I sat down on a Tuesday morning with zero ideas. Ran /daily-content-researcher and got 10 trending topics. Picked 8 of them. Ran /content-ideator and got 5 variations per topic, giving me 40 potential videos. Filtered that down to 23 that felt strong. Ran /content-scripter on all 23.

By noon I had 23 videos, each with a researched hook, a 3-beat filming card, and a CTA. All saved in Airtable. All organized by platform and format. I batch-filmed 8 of them that afternoon and scheduled them for the rest of the week.

That's not a special day. That's what happens when research, ideation, and scripting are automated. The bottleneck moves from "what do I post?" to "which of these 23 videos do I want to film first?" That's a much better problem to have.

The engine also improves over time. Every video you publish and track adds data to your Supabase analytics. Every week you run /shortform-analysis, you get better insights into what works for YOUR specific audience. The hooks get sharper. The topic selection gets more accurate. The scripts get closer to your actual voice.

Month one, the system is useful. Month three, it knows your brand better than most human assistants would.

Get the Templates

Everything in this guide is free to build from scratch. But if you want to skip the configuration work and start running the engine today, I put together a template pack with every file pre-written and ready to install.

What's in the pack:

• CLAUDE.md template - Full configuration file with every section, pre-formatted. Fill in your details and go.
• 7 skill files - daily-content-researcher, content-ideator, content-scripter, post-content, shortform-analysis, competitor-analyst, video-downloader. Each one tested and production-ready.
• 4 reference file templates - avatar.md, scripting-voice.md, viral-content-patterns.md, hook-swipe-file.md. Each with examples and fill-in-the-blank sections.
• Airtable CSV import - Creates your content pipeline with all the right columns, views, and statuses. Import it into a new base and you're set.
• Supabase migration SQL - Run it once in your Supabase SQL editor. Creates all the analytics tables the engine needs.
• Setup checklist - A printable step-by-step list so you don't miss anything during installation.

Follow Along

I post Claude Code tutorials and AI content creation walkthroughs daily. If you want to see the engine running live, watch me build new skills, or get updates when I add new features to the system, follow along on Instagram.

If you build your own content engine using this guide, send me a screenshot. I want to see it. Seriously. Nothing is more interesting to me right now than watching other people take this system and make it their own.

The tools are free. The strategy is on this page. The only thing between you and a working AI content engine is 30 minutes of setup.

Go build it.