Claude Code Quickstart Core Guide

Claude Code is Anthropic's official AI-native coding tool. It integrates large-language-model capability directly into the terminal, so you can complete programming tasks by collaborating with AI in natural language. Unlike traditional code-completion tools, Claude Code can understand the context of an entire project and execute complex development tasks. From code generation to refactoring, from debugging to documentation writing, it can handle all of them.

This chapter helps you quickly master the core usage of Claude Code, including installation and setup, basic operations, practical techniques, and commonly used commands. Whether this is your first time using an AI coding tool, or you want to use Claude Code more efficiently, you will find what you need here.

Quick Installation

Claude Code is built on Node.js, so before installation make sure Node.js 18 or above is installed on your system. The process is very simple and usually takes only a few minutes.

Why You Need Claude Code

In traditional development workflows, developers frequently switch between editor, terminal, browser, and docs. Claude Code unifies these workflows into one interface: in the same terminal window, you can write code, run tests, read docs, and even collaborate with teammates. More importantly, it can understand your project structure and remember your coding habits, becoming a true programming assistant.

Method 1: Manual Installation

Manual installation is suitable for developers who like full control over each step, and it also helps you clearly understand tool components.

# Install Claude Code CLI globally
# Use -g to install command globally, so it can be used in any directory
npm install -g @anthropic-ai/claude-code

# Verify installation
# If version is shown (for example 0.1.25), installation succeeded
claude --version

During installation, npm automatically downloads dependencies and configures environment variables. If you run into permission problems, try sudo (macOS/Linux) or run terminal as administrator (Windows).

Method 2: Let an AI Agent Install It for You

If you are already using other AI coding assistants (such as Cursor, Windsurf, or the AI Agent in this project), you can let them complete installation for you. The benefit is that AI can detect your environment automatically, handle dependency conflicts, and choose the best installation route for your system.

You can just say:

Help me install Anthropic Claude Code.

Or more specifically:

Install Claude Code CLI and check whether my Node.js version is compatible.

An AI Agent will:

1. Check current Node.js version
2. Prompt you to upgrade if requirements are not met
3. Run installation commands
4. Verify installation result
5. Try automatic fixes if there are issues

First Launch and Initialization

After installation, enter your project directory and start Claude Code:

# Enter project directory (Claude Code works in current directory)
cd /path/to/your/project

# Start Claude Code
claude

At first launch, Claude Code guides you through several important setup steps:

1. Sign in to Anthropic account: you need an Anthropic account to use Claude Code. If you do not have one, you will be prompted to register.
2. Choose a plan:
   • Free plan: suitable for personal learning and light usage, with call limits
   • Pro plan: suitable for professional developers, with higher quota and priority response
3. Accept terms: read and accept Anthropic terms and privacy policy
4. Optional: configure API key: if you have a custom key (for example from a third-party provider), configure it here

Special Note for Users in Mainland China

Due to network reasons, users in mainland China may not be able to directly access Anthropic official services. Claude Code supports third-party services compatible with Anthropic API format, and this is technically feasible.

You have two options:

1. Use API token directly: buy a token from a provider compatible with Anthropic API and configure it with environment variables
2. Use a Coding Plan: some providers offer coding-optimized plans that are usually more cost-effective for coding scenarios

Recommended approach: let an AI Agent help you configure. You only need to provide provider config information (API endpoint, key, etc.), and AI can set environment variables correctly.

See detailed setup guide: How to install claudecode and configure environment variables

Quick Start: Run a Few Small Experiments

After installation, do not rush into formal projects. Run a few small experiments first to understand how Claude Code works. These three experiments are designed from easy to advanced, corresponding to three core abilities: natural-language understanding, content generation, and code execution.

Experiment 1: Conversation - Feel AI Understanding

The purpose is to experience Claude Code's natural-language understanding. Unlike normal search engines, Claude Code can understand context, carry multi-turn conversation, and adjust answers from your feedback.

Try these prompts:

Hello, who are you?

Claude introduces itself as Claude Code, an AI coding assistant by Anthropic.

What is a closure? Give me the too-long-didnt-read version.

Observe how Claude uses "too-long-didnt-read" as a hint and gives concise but accurate explanation.

What is the difference between JavaScript and TypeScript?

This is a technical comparison question. Check whether Claude provides a structured and in-depth answer.

Experiment point: note Claude's response style. It usually gives the core conclusion first, then details. This "inverted pyramid" style is excellent for fast information retrieval.

Experiment 2: Generate a Markdown Document - Experience Content Creation

This experiment demonstrates Claude Code's content-generation capability. For developers, writing docs is often painful. Claude can quickly generate clear and complete docs from requirements.

Enter this instruction:

Write a Markdown document of commonly used Git commands.
Requirements: include command, explanation, and example.

What Claude does:

1. Analyze your requirement: common Git commands, Markdown format, and three elements (command/explanation/example)
2. Plan document structure: usually grouped by usage scenario (init, daily dev, branch workflow, remote collaboration, etc.)
3. Generate content: concise explanation and practical examples for each command
4. Format output: use Markdown syntax and proper structure

Expected output sample:

# Common Git Command Cheat Sheet

## Initialize Repository

| Command | Explanation | Example |
|------|------|------|
| `git init` | Initialize new repository | `git init my-project` |
| `git clone` | Clone remote repository | `git clone https://github.com/user/repo.git` |

...

Advanced attempts: you can add extra requirements like "add Chinese comments", "sort by frequency", "include common error handling", etc., and observe how Claude adapts output.

Experiment 3: Write and Run a Game - End-to-End Coding Workflow

This is the most challenging experiment. It demonstrates Claude Code's full workflow: understand requirement, write code, create files, run program, and handle errors. Through it, you can really feel the power of an AI coding assistant.

Enter this instruction:

Write a Snake game in Python.
Requirements:
1. Use pygame
2. Show score
3. Press ESC to exit

After writing, help me run it.

Claude executes these steps:

Step 1: Check environment

• Check whether Python is installed
• Check whether pygame is available
• Prompt installation if missing

Step 2: Write code

• Create game entry file (for example snake_game.py)
• Implement movement, food generation, collision detection
• Add score rendering
• Implement ESC exit

Step 3: Run game

• Execute Python script and launch game
• Game window pops up, use arrow keys to control snake

Step 4: Follow-up support

• If there is a bug, you can directly say "snake can pass through walls, fix it"
• If you want more features, such as "increase difficulty with score", Claude can keep modifying

Value of this experiment:

1. Verify setup: confirm Claude Code can execute code correctly
2. Experience interaction: feel collaborative development with AI
3. Build confidence: see AI complete an end-to-end runnable program

Common questions:

• Q: What if pygame is not installed?

  • A: Claude detects it and suggests pip install pygame, or you can ask Claude to install it

• Q: Terminal is occupied after game starts, what should I do?

  • A: Press ESC to quit game, or keep using Claude Code in another terminal window

• Q: Can I switch language?

  • A: Absolutely. Try "write in JavaScript", "write with HTML5 Canvas", etc.