Claude Code: A Complete Beginner's Guide (Using Domestic Models)

Start from scratch, install Claude Code—which understands natural language, writes code, and works automatically—on your computer, and use it to create your first functional product.

Getting Started

• Complete Beginners: Follow Chapters 1 to 5 in order. Once you run your first web tool, you've officially started.
• Encountering Errors: Take a screenshot and give it to Claude Code or Doubao. Describe "which step I am on and what error I see." This is 10x more efficient than Googling it yourself.

Chapter 1: Mastering Basic Concepts

If you can't understand livestreams or manuals, it's usually not because you're not smart, but because the terms aren't connected. This chapter breaks down the words that will appear repeatedly.

1.1 Three Core Concepts

1Model = Brain (Only thinks, no hands or feet)
2Chatbot = Brain + Mouth (Can talk but can't do)
3Agent = Brain + Hands/Feet + Toolbox (Can talk and do)

• Model: GPT, Claude, Gemini, DeepSeek, Doubao, Qwen, Kimi, GLM, MiniMax... these are models, the "brain."
• Chatbot: ChatGPT web, Doubao App, DeepSeek dialog box. You ask, it answers, but you have to do the work yourself.
• Agent: You say "help me build a website," and it creates folders, writes code, runs it, and debugs it on your computer, finally telling you "it's done." Claude Code is an Agent.

The same model in different tools has vastly different capabilities. The same Claude model is just a chat in a browser, but in Claude Code, it can write code itself.

1.2 Token, Context, Harness

• Token: The smallest unit AI uses to understand text. 1 Chinese character is about 1-2 Tokens; 1 English word is about 1 Token. Tokens are your "electricity bill." Don't save Tokens during the learning phase.
• Context: Everything in this round of conversation. AI has no long-term memory; it can only refer to what's on its "desk."
• Context Window: The size of each model's "desk." In 2026, mainstream models have reached 1 million Tokens (about 700k-800k words).
• Context Pollution: Chatting about different things in one window makes the AI confused. One window for one task; open a new window for a new topic.
• Harness / Context Engineering: Not a software, but a set of methods to master AI—how to organize information, manage context, guide AI, and verify results. Claude Code is great because its Harness is the most mature.

1.3 What is a Skill?

A Skill is an SOP Manual for Claude.

McDonald's Big Macs taste the same globally not because every shop has a master chef, but because every shop has an SOP. A Skill is this SOP for Claude—defining "what to do first, what to do next, what tools to use, and what counts as success."

• Write once, call repeatedly.
• Writing your own SOP is like "teaching" your work experience to Claude; it becomes your "virtual employee."

1.4 Three Payment Methods

The most realistic solution: Use domestic model APIs + cc-switch for one-click switching. Detailed in Chapter 3.

1.5 Auxiliary Tools to Know Before Installation

1.6 MCP — The "Socket Standard" for Agents

MCP stands for Model Context Protocol.

Think of it as a socket standard—allowing Agents to connect to various external tools and data sources (Feishu, databases, browsers, Xiaohongshu, WeChat, etc.).

Analogy:

• A USB port lets you plug peripherals into your computer; plug in a speaker, and the computer gains the ability to "play music."
• MCP lets you plug tools into Claude Code; plug in Tavily, and it gains the ability to "search real-time trends"; plug in xiaohongshu-mcp, and it can "post to Xiaohongshu."

By default, Claude Code only processes info you give it and won't actively search the internet. You need MCP for searching, reading databases, or calling external APIs.

1.7 Multimodal — AI Does More Than "Read Text"

Modality is the type of information. Text is one modality; images, sound, and video are others.

• Unimodal: Can only process text.
• Multimodal: Can process text, images, audio, video, files, and screens simultaneously.

Mainstream model support in 2026:

Input ≠ Output

AI being able to "see" a modality doesn't mean it can "generate" it.

• Images: Almost all mainstream models can see images; only some can draw them (GPT Image, Gemini, Doubao Seedance, Jimeng, Nano Banana, etc.).
• Video: Gemini, Doubao, and Kimi can watch video; but the LLMs don't directly make video; they use specialized tools (Wan, Seaweed Dance, Jimeng).

Model Strengths:

Practical advice for beginners: Don't just type; learn to throw screenshots, recordings, and videos directly at the AI to increase efficiency.

1.8 AI Tool Landscape (Who is Who)

One map to clear up the confusion about different tools.

Agent Tools (Help you work)
``text
Agent Tools
├── Command Line (Terminal-based, most powerful)
│ ├── Claude Code —— By Anthropic, currently strongest, the star of this guide
│ └── Codex CLI —— By OpenAI, open source
│
├── GUI (Visual interface, user-friendly)
│ ├── Codex Desktop/Web —— By OpenAI
│ ├── Claude Cowork —— By Anthropic, for non-technical users
│ ├── Trae Solo —— By ByteDance, domestic, has Code and MTC modes
│ ├── WorkBuddy —— By Tencent, domestic, for non-dev office work
│ └── Kiro —— By AWS, spec-driven IDE (VS Code fork)
│
├── Autonomous (Runs in background)
│ ├── OpenClaw —— Formerly Clawdbot, open source, remote control via WeChat/TG
│ └── Hermes —— By Nous Research, open source, persistent memory
│
└── Workflow Orchestration (Building block automation)
└── COZE —— By ByteDance, drag-and-drop workflows
``

Chatbot Tools (Talk to you)
``text
Chatbot
├── Overseas
│ ├── ChatGPT (GPT model)
│ ├── Claude.ai (Claude model)
│ ├── Gemini (Google model)
│ └── Grok (xAI/Elon Musk, for X-related tasks)
│
└── Domestic
├── Doubao (ByteDance)
├── DeepSeek (High cost-performance)
├── Qwen (Alibaba)
├── Kimi (Moonshot AI)
├── WeChat Yuanbao (Tencent, strong WeChat ecosystem data)
└── GLM / Zhipu Qingyan (Zhipu AI)
``

Common "Slang" in Comments:

1.9 One-Sentence Summary of the Relationship

1You (Human)
2  ↕ Natural Language
3Claude Code (Agent) —— Connects to external tools via MCP
4  ↕ Calls
5Model (Claude / GLM / MiniMax / Kimi…)
6  ↕ Consumes
7Token (Electricity bill)

Chapter 2: Setting Up the Environment

Install 4 things in order: Git → Node.js → VS Code → Claude Code.

2.1 Confirm Chip Type

• Mac: Apple Icon -> About This Mac. M1/M2/M3/M4 is ARM; Intel is x86_64.
• Windows: Right-click "This PC" -> Properties. Intel/AMD is x86_64.

2.2 Install Git

• Mac: Open Terminal, paste: /bin/zsh -c "$(curl -fsSL https://gitee.com/happyaicoder/HomebrewCN/raw/master/Homebrew.sh)". Verify with git --version.
• Windows: Download from https://git-scm.com/. Use default settings.

2.3 Install Node.js (18+, 20 or 22 recommended)

• Mac: Paste: /bin/bash -c "$(curl -fsSL https://gitee.com/iamzhihuix/nvm-install-cn/raw/main/install.sh)". Verify with node -v.
• Windows: Download LTS from https://nodejs.org/. If npm -v fails in PowerShell, run: Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser.

2.4 Install VS Code

Download from https://code.visualstudio.com/. On Windows, check "Add to PATH" and "Open with Code" options.

2.5 Install Claude Code

• Mac: npm install -g @anthropic-ai/claude-code --registry=https://registry.npmmirror.com. Use sudo if permission denied.
• Windows: Run PowerShell as Admin, then: npm install -g @anthropic-ai/claude-code --registry=https://registry.npmmirror.com.

2.6 Common Error Troubleshooting

Chapter 3: Configuring Claude Code with Domestic Models

Claude Code is the shell; it needs a model (brain). Official Claude models require overseas credit cards and risk bans. The most realistic solution for domestic users is using cc-switch with domestic models.

3.1 Choose a Model

Beginner Recommendation: MiniMax Coding Plan (29 RMB/month).

3.2 Install cc-switch

Download from GitHub or Gitee Mirror.

3.3 Get API Key (Example: MiniMax)

1. Register at MiniMax Platform.
2. Purchase a Coding Plan.
3. Copy the Coding Plan Key (not the regular API Key).

3.4 Configure in cc-switch

1. Open cc-switch, click the Claude icon.
2. Click +, select provider (e.g., MiniMax), paste Key, and click Add.
3. Move mouse over the model and click "Enable."

3.5 Verify

In terminal, type claude. Ask: "What model are you?" If it replies MiniMax/GLM, you're set.

Chapter 4: Claude Code Basic Operations

4.1 Two Ways to Start

• Terminal: Open a folder in VS Code, open terminal, type claude.
• Plugin: Install the "Claude Code" extension by Anthropic in VS Code.

4.2 Essential Shortcuts

4.3 Three Working Modes

Plan mode is crucial—Claude thinks through the steps before acting. Use it for complex features.

4.4 Normal vs. Deep Thinking (ultrathink)

4.5 YOLO Mode (Hands-free)

Skips permission prompts for every file change. Use: claude --dangerously-skip-permissions.

4.6 CLAUDE.md for Preferences

Write your preferences in a CLAUDE.md file in your project root (e.g., "Use Chinese for comments").

Chapter 5: Building an "Early Reading Tool"

5.1 Create Directory and PRD

Create prd.md in VS Code with requirements for an HTML/JS app that generates emoji animals when microphone volume exceeds a threshold.

5.2 Implementation

In terminal, run claude --dangerously-skip-permissions and type: @prd.md implement this feature.

5.3 Result

Open index.html in a browser, allow microphone access, and speak to see animals appear.

Chapter 6: Mastering Workflow Skills

Skills are reusable SOPs. A Skill includes a SKILL.md (config), scripts, and templates.

6.2 Using npx skills

Install skills with one command: npx skills add jimliu/baoyu-skills. Select the skill and target (Claude Code).

6.3 Skill Repositories

6.5 Creating a Custom Skill: Auto-Illustration

Use APImart's Nano Banana 2 API to create a skill that generates 3 hand-drawn style illustrations for any article you write.

6.6 Skill vs MCP vs Hooks vs Plugin

• Skill: Recipe (Steps for the chef).
• MCP: Utilities (Water/Electricity/Gas for the kitchen).

Chapter 7: Hardware and Network

• Mac is better for AI due to unified memory, but Windows works fine.
• Use TUN mode in your proxy (e.g., clash.verge) if you encounter network errors with MCP or image generation.

Appendix A: Command Cheatsheet

claude, /skills list, /clear, /compact, /doctor, /model.

Appendix B: Common Errors

Appendix C: Terminology

Final Words

Get on the road first, then talk about driving skills. Start with Chapters 2-5. AI is a 10x-100x productivity multiplier. Start today.