1. First, listen up for a moment.
If you’ve used Claude for a while, you’ve probably heard things like this.
“Using Claude Code means really using AI.”
“Have you configured CLAUDE.md?”
But when you actually search for it, you find English official documentation, developer-focused explanations, complicated installation procedures…
So you just end up chatting like ChatGPT and that’s it.
This article is written to break down that wall for people who only chat and take a step forward.
It’s not difficult. Once you understand the structure, it becomes clear.

2. Claude’s structure is literally a 3-layer building
When you first see Claude Code, it looks complicated, but actually it’s made of 3 layers.
| Layer | Role |
|---|---|
| Extension Layer (MCP, Hooks, Skills) | External connection & automation |
| Delegation Layer (Sub-agents) | Complex task sharing |
| Core Layer (Basic conversation) | Where we normally use it |
Most people only use the core layer. You ask a question in the chat window, get an answer, and done.
But power users use it like this.
They delegate complex tasks to sub-agents, connect external tools with MCP,
and only ask Claude for critical decisions.
This difference is between “just using it” and “using it properly.”

3. What is MCP, explained really simply
MCP stands for Model Context Protocol, and while the name is complex, the concept is simple.
3-1. The power drill analogy is perfect
Even if the power drill body is the same, changing the bit changes what you can do.
Drill bit → Drill holes
Abrasive bit → Sand surface
MCP is exactly that bit. Claude’s body stays the same, but connecting an MCP server dramatically expands what you can do.
3-2. MCP server examples
| MCP Server | What it enables |
|---|---|
| GitHub MCP | Issue analysis, auto PR generation |
| Notion MCP | Read/write Notion documents |
| Figma MCP | Design files directly to code |
| PostgreSQL MCP | Direct database queries |
| Blender MCP | 3D automation ← Pay attention if you use 3D! |
MCP = Claude’s ability expansion pack.
4. What about Claude Code then?
Unlike the chat Claude you use on Claude.ai website, Claude Code is Claude you use in the terminal (command line).
Claude Code → Directly touch my computer files, code, and external services
“Check ISSUE-1234 on GitHub, find related code, fix it, and submit a PR.”
It handles this all in one command. Claude decides the steps on its own.
5. Remember just 3 core concepts
5-1. CLAUDE.md — Claude’s instruction manual
CLAUDE.md is Claude’s memory about my project. If you create this file in your project folder, Claude automatically reads it when starting.
– This is a Blender automation project
– Using Python 3.11
– Save files to /outputs folder
– Write commit messages in Korean
Write it once, reuse it forever — your own Claude settings.
5-2. Connecting MCP — Just one command line
claude mcp add github — npx -y @modelcontextprotocol/server-github
# Connect Notion
claude mcp add –transport http notion https://mcp.notion.com/mcp
That’s it. After connecting, you can immediately say “Organize today’s issues from GitHub” or “Add work progress to Notion.”
5-3. Scope — Set what to share
| Scope | Meaning |
|---|---|
| local (default) | Use only in current project by me alone |
| project | Share with team members (manage with .mcp.json file) |
| user | Use across all my projects |
6. 3 real-world examples
6-1. 3D Artist (Blender + MCP)
“Unwrap UVs on the selected objects and save to the texture folder.”
Claude directly controls Blender. No need to write scripts.
6-2. Developer (GitHub + Sentry MCP)
Claude navigates from Sentry → GitHub → code files on its own while analyzing.
6-3. Automation pipeline
Discord chat arrives → Claude processes
Webhook event arrives → Claude reacts
“Using Claude as a 24/7 automation agent.”
7. Summary — This is the core. Please remember this
| Concept | One-line summary |
|---|---|
| Claude Code | Claude you use in terminal. Directly control my computer |
| MCP | Claude’s ability expansion pack. Connect external tools |
| CLAUDE.md | Claude’s project memory. Write once, reuse forever |
| Scope | Set whether I use it alone or team uses together |
| Sub-agents | Claude independently divides and handles complex tasks |
8. Next episode preview
Episode 2 is a complete breakdown of Skills that you can’t stop using.
- ✔ How Skills differ from MCP
- ✔ When to use CLAUDE.md and Skills
- ✔ 3 real-world Skills templates for blogs / Blender / automation
If this article was helpful, episode 2 on Skills is just as important as MCP. Please wait.
Let me know in the comments which MCP you’re curious about, and I’ll cover it first.
Posted on Jan 29, 2025

