[2026 Edition] The Complete Guide to Claude Code: From Installation to Advanced Techniques

Claude Code is a terminal-based AI coding agent developed by Anthropic. This article systematically covers everything you need to put it to work in a real project — from installation to practical usage.

---

What Is Claude Code

Claude Code is an agentic AI coding tool that runs in your terminal. It understands your entire project codebase and can generate, edit, and debug code — as well as handle Git operations — end-to-end based on natural-language instructions.

The key difference from traditional IDE autocomplete tools is its ability to autonomously make changes across multiple files. A single instruction like "write tests for this function" or "add error handling" can trigger coordinated edits spanning many files at once.

---

Installation and Initial Setup

System Requirements

• OS: macOS 12+, Ubuntu 20.04+, Windows (via WSL2)
• Node.js: v18 or higher
• Anthropic account: Pro plan or above (Max plan recommended)

Installation Steps

Run the following command in your terminal:

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

After installation, navigate to your project's root directory and run the claude command:

cd your-project
claude

On first launch you'll be prompted to log in with your Anthropic account. A browser window will open automatically; once authentication is complete, control returns to your terminal.

Configuring Your Project with CLAUDE.md

Creating a CLAUDE.md file in your project root lets you pass project-specific rules and context to Claude Code:

# Project Rules
- Write in TypeScript with strict mode enabled
- Use Vitest for tests
- Write commit messages in Japanese
- Match the existing code style

This file is loaded automatically at the start of every session, so you never have to repeat the same instructions.

---

Basic Usage

Code Generation and Editing

The core of Claude Code is simple: just give instructions in plain language:

• "Create an authentication middleware"
• "Add error handling to this function"
• "Look at the database schema and generate a CRUD API"

When Claude Code receives an instruction, it automatically reads the relevant files and proposes changes. Because you can review changes before they are applied, there is no risk of unintended modifications.

Git Operations

You can also manage your Git workflow with natural language:

• "Commit the changes" → automatically generates an appropriate commit message
• "Create a PR" → handles everything from branch creation to pull request submission
• "Show the diff against the main branch"

Debugging

Resolving errors is another strong suit:

• Paste an error message and say "fix this error"
• "Find out why the tests are failing"
• "Resolve the build error"

Claude Code analyzes the error message, reads the relevant code, and proposes a fix.

---

5 Practical Tips

1. Use Plan Mode to Design Large Tasks

Before tackling a complex task, use the /plan command to build an implementation plan. Claude Code will organize the files that need to change and the steps required. Reviewing and refining the plan before starting implementation dramatically reduces rework.

2. Automate Repetitive Work with Custom Commands

Create markdown files in the .claude/commands/ directory to define custom slash commands:

<!-- .claude/commands/test.md -->
Please create unit tests for this file.
Cover happy paths, error paths, and edge cases, following the existing test patterns.

Typing /test will produce consistent, high-quality tests every time.

3. Grow Your CLAUDE.md

Add rules and conventions you discover while working on the project to CLAUDE.md. Every team member benefits automatically.

4. Be Mindful of Context

Claude Code retains the conversation history within a session. You can leverage prior context with instructions like "add validation to the function we just discussed." When switching to a new task, however, running /clear to reset the context tends to improve accuracy.

5. Use Sub-agents for Parallel Processing

Claude Code internally uses sub-agents to investigate and modify multiple files in parallel. Even broad instructions like "add rate limiting to all API endpoints under src" are handled efficiently.

---

Pricing Plans and Cost Estimates

Claude Code runs on Anthropic's Claude subscription plans:

For everyday development, Max 5x ($100/month) offers the best balance. If you work on large-scale refactors or multiple projects, consider Max 20x.

For API access, the Claude Sonnet model costs approximately $3 per million input tokens and $15 per million output tokens.

---

Comparison with Other AI Coding Tools

Claude Code | Terminal-first agentic tool

Understands the entire project and autonomously makes cross-file changes. Lightweight and fast as a CLI tool. Highly extensible via CLAUDE.md and custom commands.

Learn more

Cursor | AI-powered IDE editor

An AI editor built on VS Code. Excels at intuitive, GUI-driven workflows. Best suited for work that stays within the editor, such as code completion and inline editing.

Learn more

Devin | Autonomous AI software engineer

A fully autonomous AI engineer developed by Cognition. Hand it a task and it independently handles everything from design to implementation. Requires minimal human supervision.

Learn more

---

Frequently Asked Questions (FAQ)

Q. Is Claude Code free to use?

It requires a Claude Pro plan ($20/month) or higher. It is not available on the free plan.

Q. Does it work on Windows?

Yes, via WSL2 (Windows Subsystem for Linux). Install Node.js inside WSL2 and run npm install -g @anthropic-ai/claude-code as usual.

Q. Can it be used offline?

No. Claude Code communicates with Anthropic's API and requires an internet connection.

Q. Is my private repository code safe?

Code accessed by Claude Code is sent to Anthropic's API, but per Anthropic's privacy policy it is not used to train models.

Q. Which programming languages are supported?

Major languages including TypeScript, Python, Rust, Go, and Java are supported. TypeScript and Python tend to yield the highest accuracy.

---

Related: securing the skills Claude Code assumes

Claude Code is an agent that takes the terminal as its ground. Reading the diff it proposes, interpreting an error and recovering a broken state all happen at the CLI. How to build that foundation for free is covered in "CLI and programming study in the age of AI coding agents" (WebTerm and WebTerm Learn are both free, and Progate has a Claude Code Introduction (beta) in Japanese).