Claude Code: Complete Guide for Developers

Claude Code is Anthropic’s agentic coding tool. Instead of suggesting a line of code and waiting, it reads  your codebase, edits files across your project, runs commands, and checks its own work, all from a plain English instruction. This guide covers what Claude Code is, how to install it on every platform, what it  costs, and the features and habits that separate developers who get real leverage from it from those  who don’t. 

In this guide: 

1. What is Claude Code? 

2. Where Claude Code runs 

3. How to install Claude Code 

4. Pricing and who can use it 

5. Your first session 

6. CLAUDE.md: give Claude permanent context 

7. Hooks, skills, subagents, MCP and plugins 

8. Best practices that actually matter 

9. Claude Code in automation and CI 

10. Common mistakes 

11. Claude Code vs Cursor and other tools 

12. FAQ

What is Claude Code? 

Claude Code is an AI coding assistant that can plan a change, write code across several files, run your  tests, and fix what breaks. Anthropic describes it as an agentic coding tool that reads your codebase,  edits files, runs commands, and integrates with your development tools. 

The word that matters is agentic. A chat assistant answers and stops. Claude Code works in a loop: it  explores the project, forms a plan, makes edits, runs a command to check the result, reads the output,  and keeps going until the job is done or it needs your input. You can watch, redirect, or step away. 

Typical jobs include: 

• Building a feature from a description and wiring it across several files 

• Tracing a bug from an error message to its root cause and fixing it 

• Writing tests for untested code, then running them and fixing failures 

• Resolving merge conflicts, fixing lint errors, updating dependencies 

• Creating commits and opening pull requests through git 

• Explaining an unfamiliar codebase, as you would ask a senior engineer 

Where Claude Code runs 

Claude Code is not only a terminal tool. It runs on several surfaces that share the same engine, so your  project’s CLAUDE.md files, settings and MCP servers carry across all of them. 

How to install Claude Code 

Anthropic recommends the native installer. Open a terminal and run the command for your system.

macOS, Linux and WSL: 

curl -fsSL https://claude.ai/install.sh | bash 

Windows PowerShell: 

irm https://claude.ai/install.ps1 | iex

Windows CMD: 

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd Package-manager options also exist: 

# Homebrew (macOS/Linux) 

brew install --cask claude-code 

# WinGet (Windows) 

winget install Anthropic.ClaudeCode 

You can also install through apt, dnf or apk on Debian, Fedora, RHEL and Alpine.

Good to know: 

• Native installs update themselves in the background. Homebrew and WinGet installs do not, so  run brew upgrade claude-code or winget upgrade Anthropic.ClaudeCode now and then. • Homebrew has two casks: claude-code follows the stable channel (typically about a week  behind) and claude-code@latest gets releases as soon as they ship. 

• On native Windows, Git for Windows is recommended so Claude Code can use its Bash tool.  Without it, Claude Code falls back to PowerShell. You can also run it inside WSL. • Many older tutorials show an npm install. Check the official setup page for the current  recommended method before following them. 

Verify the install: 

claude --version 

claude doctor

If your shell says claude is not found, the install directory is not on your PATH yet. Open a new terminal  window first, then fix your PATH if the problem remains. 

Start it: 

cd your-project 

claude

On first run you’ll be asked to log in. If you have an ANTHROPIC_API_KEY environment variable set, Claude  Code skips the login prompt and asks you to approve the key instead. 

Claude Code pricing and who can use it 

Claude Code needs a paid account. The free Claude.ai plan does not include it. You can use it through: 

• Claude Pro: $20 per month ($17 per month billed annually) 

• Claude Max 5x: $100 per month, roughly five times Pro capacity 

• Claude Max 20x: $200 per month, roughly twenty times Pro capacity 

• Team or Enterprise plans: seat-based, with Claude Code access depending on seat type

• Anthropic Console / API: pay per token instead of a flat subscription

Pro is a sensible place to start. Developers who hit usage limits regularly, or who run several agents for  hours, usually move to a Max plan. Plans and limits change, so confirm current prices on the official  pricing page before you subscribe. 

Your first session 

Start in a project folder and talk to Claude Code the way you’d brief a teammate: explain how authentication works in this project 

add input validation to the signup form, write tests for it, run them and fix any failures commit my changes with a descriptive message 

A few habits help immediately: 

• Reference files with @, for example @src/auth/login.ts, so Claude reads the file instead of  guessing. 

• Paste screenshots or error messages directly into the prompt. 

• Press Esc to stop Claude mid-action without losing context. 

• Press Esc twice or run /rewind to restore an earlier conversation and code checkpoint.

• Run /clear when you switch to an unrelated task. 

CLAUDE.md: give Claude permanent context 

CLAUDE.md is a markdown file in your project root that Claude Code reads at the start of every session.  Use it for things Claude cannot infer from the code: build and test commands, style rules that differ from  defaults, branch and pull request conventions, architectural decisions, and environment quirks. 

Run /init to generate a starter file from your project structure, then refine it. 

# Code style - Use ES modules (import/export), not CommonJS (require) 

# Workflow 

- Typecheck after a series of code changes 
- Prefer running a single test file over the full suite

The most common mistake is making it too long. Anthropic’s guidance is to keep it short and cut any line  whose removal would not cause a mistake. A bloated file makes Claude ignore the rules that matter.  Run /context to confirm the file loaded. Claude Code also builds auto memory as it works, saving what it  learns across sessions. 

Hooks, skills, subagents, MCP and plugins 

These five features customize Claude Code. Knowing which one fits which problem saves a lot of time. Hooks: guaranteed automation 

Hooks are shell commands, HTTP calls, prompts or subagents that run automatically at set points in  Claude’s workflow. Unlike instructions in CLAUDE.md, which are advice, hooks are deterministic, so use  them for anything that must happen every single time.

Common events include SessionStart, UserPromptSubmit, PreToolUse (before a tool call, and it can block  it), PostToolUse (after a call), and Stop (when Claude finishes). You configure them in  .claude/settings.json for a project or ~/.claude/settings.json for all your projects. 

This example runs a linter after every file edit: 

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            "type": "command",
            "command": "/path/to/lint-check.sh"
          }
        ]
      }
    ]
  }
}

For policy hooks that block actions, remember that exit code 2 is what blocks. A plain exit code 1 is  treated as a non-blocking error and the action goes ahead. Run /hooks to browse what is configured. 

Skills: reusable know-how 

A skill is a folder containing a SKILL.md file, placed in .claude/skills/. Claude applies a skill when it is  relevant, or you invoke it directly, for example /fix-issue 1234. Use skills for domain knowledge or  workflows that matter only sometimes, so they don’t bloat every session the way a long CLAUDE.md would. 

Subagents: isolated helpers 

Subagents run in their own context window with their own allowed tools and report back a summary.  Define them as markdown files in .claude/agents/, or just tell Claude: “use a subagent to review this  code for security issues.” They are ideal for research that reads many files, and for independent review  of finished work. 

MCP: connect your tools 

The Model Context Protocol is an open standard for connecting AI tools to outside data and services.  With MCP servers, Claude Code can read design docs, update tickets in Jira, query databases, pull data  from Slack, or use your own tooling. Add a server with: 

claude mcp add –transport http notion https://mcp.notion.com/mcp 

Plugins: bundles of all of the above 

Run /plugin to browse the marketplace. A plugin packages skills, hooks, subagents and MCP servers into  one installable unit. 

Best practices that actually matter 

Anthropic’s own guidance rests on one constraint: Claude’s context window fills up fast, and  performance drops as it fills. Almost every good habit is a way of protecting it.

1. Give Claude a way to verify its work. This is the highest-leverage habit. Provide tests, a build  command, a linter or a screenshot comparison so Claude can check itself and iterate. Without a check,  “looks done” is the only signal, and you become the verification loop. 

2. Explore, plan, then code. Use plan mode (press Shift+Tab until plan mode shows, or start with claude  –permission-mode plan) so Claude reads and plans without changing anything. Then implement, then  commit. Skip planning when you could describe the diff in one sentence. 

3. Be specific. “Add tests for foo.py” works worse than “write a test for foo.py covering the logged-out  case, avoiding mocks.” Point to existing patterns in your codebase and name the files. 

4. Manage context aggressively. Use /clear between unrelated tasks, /compact with instructions when  you want a controlled summary, and subagents for investigation so the reading doesn’t clutter your  main session. 

5. Course-correct early. If you’ve corrected Claude twice on the same issue, the context is cluttered with  failed attempts. Run /clear and rewrite the prompt with what you learned. 

6. Configure permissions sensibly. Claude Code asks before actions that change your system. Pre approve safe commands with /permissions and use /sandbox for OS-level isolation. Auto mode uses a  separate classifier to review actions instead of asking you each time. Understand the mode you’re in  before you leave a session unattended. 

7. Let Claude interview you for big features. Ask it to question you about the hard parts, then write a  spec to a file. Start a fresh session to implement from that spec. 

One caveat: checkpoints only track changes made through Claude’s file-editing tools, not changes made  by Bash commands. Checkpoints are not a replacement for git. 

Claude Code in automation and CI 

Claude Code follows the Unix philosophy, so it composes with other tools. Non-interactive mode runs  with -p: 

claude -p “Explain what this project does” 

claude -p “List all API endpoints” –output-format json 

git diff main –name-only | claude -p “review these changed files for security issues” 

You can fan work out across many files by looping over claude -p, or use /batch to split a change across  multiple subagents, each in its own git worktree. For CI, GitHub Actions and GitLab CI/CD integrations  automate pull request review and issue triage. You can also schedule recurring work, such as morning  PR reviews or weekly dependency audits, using routines or desktop scheduled tasks. 

To resume work later, use claude –continue or claude –resume. 

If you want to build your own agents on the same foundation, the Claude Agent SDK (formerly the  Claude Code SDK) exposes Claude Code’s tools and agent loop in Python and TypeScript. 

Five common mistakes 

1. The kitchen-sink session. Mixing unrelated tasks fills the context. Fix: /clear between tasks.

2. Correcting again and again. Failed attempts pollute the context. Fix: restart with a better  prompt after two misses.

3. An over-long CLAUDE.md. Important rules get lost. Fix: prune ruthlessly, or turn rules into  hooks. 

4. Trusting without verifying. Plausible code can miss edge cases. Fix: always provide tests or a  check. 

5. Unscoped investigation. “Look into this” can make Claude read hundreds of files. Fix: scope it,  or use a subagent. 

Claude Code vs Cursor and other AI coding tools 

Cursor is a standalone AI-first editor, while Claude Code is agent-first and works in your terminal, IDE,  desktop app and browser. Many developers use both: an editor for hands-on work and Claude Code for  larger multi-file tasks and automation. Because Claude Code has a VS Code extension that also installs  into Cursor, the two are not mutually exclusive. The better choice depends on whether you want the AI  inside your editor or running tasks alongside it. 

Conclusion 

Claude Code rewards a different way of working. Describe the outcome, give Claude a way to verify it,  protect the context window, and encode your team’s rules once in CLAUDE.md, hooks and skills. Start  with a small task in a real project, add one customization at a time, and let the results tell you where to  invest next. 

Build with easycomm.io. easycomm.io designs and builds custom software and AI solutions, including AI  CCTV, AI hybrid solutions, robotics (ROS-based tech stack), web development, app development, AI  development, real estate software with a built-in CRM, custom software development, an LMS with  AR/VR features, ERP and CRM systems. Have a project in mind? Get in touch with the easycomm.io team  at easycomm.io to talk through your idea. 

Sources 

• Claude Code overview: https://code.claude.com/docs/en/overview 

• Best practices for Claude Code: https://code.claude.com/docs/en/best-practices • Hooks reference: https://code.claude.com/docs/en/hooks 

• Claude Code setup: https://code.claude.com/docs/en/setup 

• Claude pricing: https://claude.com/pricing