Skip to content
FrankX.AI
Creator SystemsJan 21, 20267 min read1,378 words

Getting Started with Agentic Creator OS

TL;DR

Set up your own AI operating system in 30 minutes. Step-by-step guide to installing departments, MCP servers, and running your first automated workflow.

Frank Riemer
FrankX
AI Architect & Independent Creator
Ex-Oracle AI Architect · Starlight & ACOS Systems
Set up your own AI operating system in 30 minutes. Step-by-step guide to installing departments, MCP servers, and running your first automated workflow.
Reading Goal

Set up your own Agentic Creator OS and run your first automated workflow.

Getting Started with Agentic Creator OS

The same system running my website, my music production, and my enterprise projects.

TL;DR

Clone the repo. Run the installer. Build the MCP servers. Configure Claude Code. Create your CLAUDE.md context file. Takes 30-60 minutes. By the end, you have 5 AI departments—Content, Dev, Design, Marketing, Business—working together on your projects. This is the exact setup I use to manage 12,000+ Suno songs, a Next.js website, and enterprise consulting.

This guide walks you through setting up the Agentic Creator OS from scratch. I built this after realizing my Claude Code sessions were getting too fragmented—different contexts, different skills, no coherent system. Now everything lives in one place.

Prerequisites:

  • Basic command line familiarity (cd, ls, npm)
  • Node.js 18+ installed
  • Claude Code CLI installed (get it here)
  • 30-60 minutes of focused time

Let's build.

What You're Building

The Agentic Creator OS transforms Claude Code from a chat tool into a full operating system with:

What You're Building diagram 1
What You're Building

Think of it as having 5 AI employees that work together to run your projects.

Step 1: Get the Code

First, clone the repository:

# Clone the repo
git clone https://github.com/frankxai/agentic-creator-os.git

# Navigate into it
cd agentic-creator-os

# Look around
ls -la

You should see:

  • departments/ - The 5 AI department configurations
  • mcp-servers/ - Local tools for Claude Code
  • skills/ - Workflow definitions
  • workflows/ - Automation templates
  • install.sh - Setup script

Step 2: Run the Installer

The installer copies configurations to your Claude Code setup:

# Make the script executable
chmod +x install.sh

# Run it
./install.sh

This installs:

  • Skills to ~/.claude/skills/
  • Agent configs to ~/.claude/agents/
  • Workflow templates to ~/.claude/workflows/

Verification: Check that files were copied:

ls ~/.claude/skills/
# Should show: agentic-creator-os, content-department, etc.

Step 3: Build the MCP Servers

MCP (Model Context Protocol) servers give Claude Code local superpowers. Let's build them:

# Install dependencies
npm install

# Build all servers
npm run build:all

This compiles 6 MCP servers:

  1. filesystem - Read, write, search files
  2. database - Local SQLite operations
  3. browser - Automated web actions
  4. website - Deploy to Vercel/Netlify
  5. email - Send notifications
  6. creator - Business tools (CRM, invoicing)

Note: Building may take 2-5 minutes depending on your machine.

Step 4: Configure Claude Code

Now we need to tell Claude Code about our new servers. Open your Claude Code settings:

# Open Claude Code settings
claude config

Or edit directly at ~/.config/claude/settings.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "node",
      "args": [
        "path/to/agentic-creator-os/mcp-servers/filesystem-mcp/build/index.js"
      ]
    },
    "database": {
      "command": "node",
      "args": [
        "path/to/agentic-creator-os/mcp-servers/database-mcp/build/index.js"
      ]
    }
    // Add other servers as needed
  }
}

Replace path/to/ with your actual installation path.

Step 5: Create Your CLAUDE.md

This is where the magic happens. The CLAUDE.md file is your AI's context - everything it needs to know about YOU.

Create a file in your project root:

touch CLAUDE.md

Add your context:

# My Intelligence System

## Who I Am

I'm a [your role] building [your project].

## Current Project

- Name: [Project Name]
- Goal: [What you're building]
- Stack: [Your technologies]

## My Preferences

- I prefer [X] over [Y]
- Always [do this thing]
- Never [do that thing]

## Active Tasks

- [ ] [Task 1]
- [ ] [Task 2]

Pro tip: The more context you add, the smarter your AI becomes. Update this file regularly.

Step 6: Test Your First Department

Let's test the Content department. Open Claude Code in your project:

claude

Then try:

> Help me write a blog post outline about [your topic]

Claude should:

  1. Read your CLAUDE.md for context
  2. Use your preferences
  3. Generate content in your style

Try more commands:

  • "Create a Twitter thread about [topic]"
  • "Draft an email to [recipient]"
  • "Write documentation for [feature]"

Step 7: Run a Workflow

Now let's run a complete workflow. If you have the /publish skill installed:

> /publish "My First Blog Post"

This triggers:

  1. Content department writes the post
  2. Design department creates a hero image (if configured)
  3. Dev department deploys to your site
  4. Marketing department generates social posts

Note: Some workflows require additional configuration (API keys, deployment settings).

Step 8: Customize Your Setup

The real power comes from customization. Here's what to explore:

Add New Skills

Create custom skills in ~/.claude/skills/:

# /my-custom-workflow

When activated, do:

1. [Step 1]
2. [Step 2]
3. [Step 3]

Modify Departments

Edit department configs in departments/ to match your workflow.

Create Templates

Add templates for common tasks in workflows/.

Common Issues & Solutions

"MCP server not found"

  • Check that the path in settings.json is correct
  • Verify the server was built successfully
  • Try rebuilding with npm run build:all

"Command not recognized"

  • Ensure skills were installed to ~/.claude/skills/
  • Re-run ./install.sh
  • Check Claude Code is reading from the right config

"Context seems wrong"

  • Verify CLAUDE.md exists in your project root
  • Check file isn't too large (keep under 10KB)
  • Update context with current project state

What's Next?

Now that you have the basics working:

Level 1: Daily Usage

  • Use Content department for writing tasks
  • Use Dev department for coding help
  • Build your CLAUDE.md over time

Level 2: Workflow Creation

  • Create custom skills for your repeated tasks
  • Set up automation triggers
  • Connect external services

Level 3: Full Automation

  • Deploy MCP servers for email, browser automation
  • Build end-to-end workflows
  • Share your setup with the community

Resources

Official

Tutorials

Community

Recap

In this guide, you:

  1. ✅ Cloned the Agentic Creator OS repository
  2. ✅ Ran the installer to copy skills and configs
  3. ✅ Built the MCP servers for local tools
  4. ✅ Configured Claude Code to use the servers
  5. ✅ Created your CLAUDE.md context file
  6. ✅ Tested your first department
  7. ✅ Ran a complete workflow
  8. ✅ Learned how to customize

Total time: ~30-60 minutes

Your Turn

The best way to learn is to use it. Pick ONE thing you need to do today and try it with your new AI OS:

  • Write that blog post you've been putting off
  • Fix that bug that's been annoying you
  • Draft that email you keep avoiding

Start small. Build the habit. Watch your productivity compound.

FAQ

How long does setup take?

30-60 minutes for basic installation. Another hour or two to customize CLAUDE.md and create your first custom skills. The system becomes more valuable the more context you feed it.

Do I need coding experience?

Basic command line familiarity is required (cd, npm, git clone). You don't need to write code—just follow the steps. The MCP servers come pre-built.

What's the difference between this and just using Claude Code?

Raw Claude Code starts fresh each session. Agentic Creator OS gives it persistent context about you, your projects, and your preferences. It's the difference between a smart assistant and one that actually knows your work.

Can I use this for a team?

The current version is designed for solo creators. Team features (shared contexts, role-based access) are on the roadmap. For now, each team member would run their own instance.

What if I get stuck?

Open an issue on GitHub. The community is active and I respond to questions directly.

Related:

Stay in the intelligence loop

Weekly field notes on AI systems, production patterns, and builder strategy.

Occasional FrankX field notes. Unsubscribe anytime. Privacy details.