Claude Code Complete Guide: AI-Powered Terminal Coding in 2026

What is Claude Code?

Claude Code is an agentic coding tool from Anthropic that operates directly in your terminal. Unlike autocomplete tools that suggest the next line, Claude Code reads your entire codebase, plans multi-step changes, writes code across files, runs tests, and commits to git — all from natural language instructions.

It has accumulated over 100,000 GitHub stars since its general availability release and is used daily by tens of thousands of developers at companies including ServiceNow (29,000+ developers) and Allianz.

The tool is available on multiple surfaces: a terminal CLI, a VS Code extension, a JetBrains plugin, a dedicated desktop app with visual diff review, a web version at claude.ai/code, and a Slack integration.

How It Works: The Agentic Loop

Claude Code follows a continuous explore → plan → execute → verify loop:

flowchart LR
  A[Explore Codebase] --> B[Plan Changes]
  B --> C[Execute: Edit / Bash / Git]
  C --> D[Verify: Tests / Lint / Diff]
  D -->|Passes| E[Commit & PR]
  D -->|Fails| A

At each step, Claude can read files, search with glob patterns, run shell commands, and edit source files. It requests permission before modifying your system by default, and you can configure progressively less oversight as trust builds.

Key Specifications

Installation

Prerequisites

• Node.js 18+ (for npm-based install) or a compatible system.
• An Anthropic account with an active subscription (Pro, Max, Team, or Enterprise) or API access.

macOS / Linux (Recommended)

Anthropic now recommends the native installer — the npm package is deprecated:

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

macOS via Homebrew

brew install --cask claude-code

Windows

# PowerShell (recommended)
irm https://claude.ai/install.ps1 | iex

# WinGet
winget install Anthropic.ClaudeCode

npm (Deprecated, Migration Available)

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

If you previously installed via npm, migrate to the native version:

claude install

Verify Installation

claude --version

First Run

Navigate to any project directory and start a session:

cd your-project
claude

On first launch, Claude Code guides you through browser-based authentication. After that, you are ready to work.

Core Slash Commands

Once inside an interactive session, these slash commands are essential:

CLI Flags

Use flags for non-interactive or scripted workflows:

# Quick query, plain text output
claude -p "Explain the structure of this project"

# Structured output for scripting
claude -p "List all API endpoints" --output-format json

# Streaming JSON for real-time processing
claude -p "Analyze this log" --output-format stream-json

# Specify a model
claude --model claude-sonnet-4-6

# Continue last session
claude --continue

# Resume from a list of sessions
claude --resume

# Skip permission prompts (use with caution)
claude --dangerously-skip-permissions

# Auto mode — classifier handles approval
claude --permission-mode auto -p "fix all lint errors"

CLAUDE.md: The Project Memory System

The CLAUDE.md file is one of the most impactful features. Claude reads it at the start of every session to understand your project conventions, build commands, and preferences.

Run /init to generate a starter file based on your codebase, then refine it:

# Code style
- Use ES modules (import/export), not CommonJS
- TypeScript strict mode enabled
- Destructure imports when possible

# Workflow
- Run `npm run typecheck` after any type changes
- Prefer `npm test -- --run` for single-file tests
- Commit messages follow Conventional Commits

CLAUDE.md File Locations

Claude Code checks multiple locations, applied hierarchically:

What to Include vs. Exclude

Keep it concise. If there are too many rules, Claude will ignore them. A good test: remove any line where Claude already does the right thing without it.

Plan Mode

Plan mode forces Claude to think before acting. It enters a read-only state — no file writes, no command execution — and focuses on analysis:

plan: I want to add Google OAuth. What files need to change?
plan: What's the session flow? Create a detailed implementation plan.

The recommended four-phase workflow:

1. Explore — Enter plan mode. Read relevant files, understand existing patterns.
2. Plan — Ask Claude to create a step-by-step implementation plan. Press Ctrl+G to open it in your editor for direct editing.
3. Implement — Switch out of plan mode. Let Claude code, verifying against the plan.
4. Commit — Ask Claude to commit with a descriptive message and open a PR.

Activate plan mode with Shift+Tab twice, Alt+M, or the /plan command.

MCP Integration

The Model Context Protocol (MCP) is Anthropic’s open standard for connecting AI models to external tools and data sources. With MCP, Claude Code can query databases, post to Slack, create GitHub issues, read Jira tickets, and interact with any service that exposes an MCP server.

Configure MCP servers via CLI or in .claude/settings.json:

# Add an MCP server
claude mcp add github --transport http

# List connected servers
claude mcp list

Project-level MCP servers are configured in .mcp.json at your project root, allowing the team to share integrations checked into version control.

Popular MCP Servers

Extending Claude Code

Beyond MCP, Claude Code offers several extension mechanisms:

Hooks

Hooks run scripts automatically at specific points in the workflow — for example, auto-formatting after every file edit:

Write a hook that runs Prettier after every file edit.

Claude can write hooks for you. Run /hooks to see what is configured.

Skills

Skills are reusable instruction files stored in .claude/skills/. They give Claude domain knowledge on demand:

Create .claude/skills/api-conventions/SKILL.md:

---
name: api-conventions
description: REST API design conventions
---

# API Conventions
- Use kebab-case for URL paths
- Use camelCase for JSON properties
- Always include pagination for list endpoints
- Version APIs in the URL path (/v1/, /v2/)

Skills load automatically when the topic is relevant, or you can invoke them with /skill-name.

Sub-agents

Sub-agents run in isolated context windows with their own tool permissions. They are useful for investigation tasks that would otherwise fill your main context:

Use a subagent to investigate how our authentication system handles token refresh.

Define custom sub-agents in .claude/agents/:

---
name: security-reviewer
description: Reviews code for security vulnerabilities
tools: Read, Grep, Glob, Bash
model: opus
---

You are a senior security engineer. Review code for:
- Injection vulnerabilities (SQL, XSS, command injection)
- Authentication and authorization flaws
- Secrets or credentials in code

Plugins

Run /plugin to browse the community and official plugin marketplace. Plugins bundle skills, hooks, sub-agents, and MCP servers into a single installable unit.

Multi-Agent and Parallel Workflows

Claude Code scales horizontally. Beyond a single session, you can run multiple Claudes in parallel:

Git Worktrees

Use git worktrees to run isolated Claude sessions on separate branches simultaneously:

git worktree add ../feature-branch feature-branch
cd ../feature-branch && claude

Each worktree has its own filesystem and git history — sessions never collide.

Agent Teams

Agent teams provide automated coordination of multiple Claude sessions with shared tasks and messaging. A lead agent decomposes work, dispatches sub-tasks to specialized agents, and integrates the results.

Desktop App Sessions

The Claude Code desktop app manages multiple local sessions visually, each in its own worktree, with split-pane diff review.

Claude Code on the Web

The browser-based version runs tasks on Anthropic-managed cloud infrastructure in isolated sandboxes. Connect GitHub repositories, delegate multiple tasks in parallel, and track progress in real-time.

Security and Permissions

Claude Code runs locally and communicates directly with model APIs — no backend server or remote code index. It asks for permission before modifying files or running commands.

Three permission modes:

Configure in .claude/settings.json:

{
  "permissions": {
    "allow": ["Read", "Write(src/)", "Bash(git *)", "Bash(npm *)"],
    "deny": ["Read(.env)", "Bash(rm *)", "Bash(sudo *)"]
  }
}

IDE Integration

Claude Code works alongside your editor without requiring you to change your workflow:

VS Code

Install the extension from the marketplace. It provides inline diffs, plan review, and conversation history.

code --install-extension anthropic.claude-code

JetBrains IDEs

Available for IntelliJ IDEA, PyCharm, WebStorm, and the rest of the JetBrains ecosystem. Search “Claude Code” in the plugin marketplace.

Vim / Neovim

" In .vimrc
command! Claude claude --file %

Desktop App

The standalone desktop app for macOS and Windows provides visual project management, parallel sessions, integrated terminal, and visual diff review.

Pricing

Claude Code is available through several plans:

When used with a Claude Console (API) account, Claude Code consumes tokens at standard API rates:

First-time users typically receive free credits to evaluate the tool.

Best Practices

Be Specific

# Good — provides constraints and context
claude "Create a Python function that accepts a list of integers and returns the sum of even numbers, using only built-in modules"

# Poor — vague, invites misinterpretation
claude "do some math"

Provide Verification Criteria

Claude performs dramatically better when it can check its own work. Include tests, expected outputs, or screenshots:

claude "Write a validateEmail function. Test cases: [email protected] → true, invalid → false, [email protected] → false. Run tests after implementing."

Use Sub-agents for Investigation

Research tasks consume context. Delegate them to sub-agents to keep your main conversation clean:

Use a subagent to investigate how we handle database migrations. Report back with a summary.

Manage Context Aggressively

The context window is the most constrained resource. Run /clear between unrelated tasks. Use /compact to condense long conversations. Avoid marathon sessions — prefer focused 5-10 turn interactions.

Leverage Auto Mode

For routine tasks like lint fixes or test repairs, auto mode eliminates approval fatigue:

claude --permission-mode auto -p "fix all lint errors in src/"

Use the Explore-Plan-Code Loop

Never jump straight to implementation on complex tasks. Use plan mode to understand the codebase first, create a plan, then execute. This prevents solving the wrong problem.

Common Mistakes

Resources

• Claude Code Official Docs — Installation, configuration, and feature reference
• Claude Code Best Practices — Official guide from Anthropic’s engineering teams
• Anthropic Console — API key management and billing
• MCP Specification — Model Context Protocol documentation and server registry
• Claude Code on GitHub — Official repository, 100K+ stars