What Is Claude Code? The Complete Guide to AI-Powered Development

What Claude Code is and how it differs from other AI coding tools

Claude Code is Anthropic's command-line interface for agentic software development. The key word is "agentic" — unlike traditional AI coding assistants that respond to individual queries or predict the next line of code, Claude Code takes on complete development tasks. Tell it "add authentication to the user API endpoints" or "refactor the payment module to use the strategy pattern" and it will read the relevant files, understand the existing code structure, plan the changes, write the code, and apply the modifications — all in a single interaction.

This distinction between agentic and assistive AI coding is fundamental. Assistive tools (like traditional Copilot) work alongside you while you code, suggesting completions and answering questions. Agentic tools (like Claude Code) work instead of you on defined tasks, executing the full cycle from understanding to implementation while you review and direct.

Claude Code runs in your terminal, not in an editor. You invoke it from your project directory, and it automatically maps your codebase — understanding file relationships, import chains, configuration files, and project conventions. It then uses this understanding to make informed changes that respect your project's patterns and architecture.

The underlying model is Claude — the same AI that powers the conversational Claude interface — but with additional capabilities for code understanding, file manipulation, and command execution. This means Claude Code inherits Claude's strengths: strong reasoning, careful analysis, high-quality output, and the 1M-token context window that can process very large codebases.

For development teams, the practical impact is significant. Tasks that previously required hours of manual work — cross-codebase refactoring, comprehensive test writing, documentation generation, architectural migration — can be completed in minutes with appropriate oversight. The human developer's role shifts from writing every line of code to directing, reviewing, and guiding the AI agent. For a deeper exploration of AI-powered development workflows, see the [Coding section](/coding) of the curriculum.

Installation and getting started

Getting started with Claude Code takes under five minutes. The tool is distributed as an npm package and requires Node.js 18 or later.

Install globally with: "npm install -g @anthropic-ai/claude-code". Once installed, navigate to your project directory in the terminal and run "claude" to start an interactive session. On first run, Claude Code will ask you to authenticate with your Anthropic account — you need a Claude Pro, Team, or Enterprise subscription, or API credits.

Claude Code immediately begins mapping your project structure. It reads your files, understands the directory layout, identifies the language and framework, and builds an internal representation of your codebase. This initial mapping happens quickly for most projects (seconds for small projects, a minute or two for very large ones) and is what enables Claude Code to make contextually aware changes rather than generic suggestions.

Your first interaction should be something concrete. Try: "Explain the architecture of this project" to verify that Claude Code has correctly understood your codebase. Then try a real task: "Add input validation to the user registration endpoint" or "Write unit tests for the authentication module." Watch how it navigates files, reads related code for context, and produces changes that fit your existing patterns.

Claude Code operates in your actual file system — changes are applied directly to your files. This means you should always work on a Git branch when using Claude Code, so you can review changes before committing and easily revert if needed. Claude Code is aware of Git and can be instructed to create commits, branch, and stage changes.

The tool supports several invocation modes: interactive mode (conversational, for exploratory tasks), one-shot mode (pass a single instruction via the command line for scripted use), and pipe mode (for integration with other tools and scripts). Most developers start with interactive mode and adopt the other modes as they build Claude Code into their workflows.

CLAUDE.md: the project knowledge file

The CLAUDE.md file is Claude Code's most distinctive feature and arguably its most powerful one. It is a markdown file placed in your project root that contains project-specific instructions, conventions, and context that Claude Code reads automatically at the start of every session.

Think of CLAUDE.md as a persistent brief for your AI developer. It replaces the need to repeat project-specific instructions in every conversation and ensures consistency across team members using Claude Code on the same project. A well-written CLAUDE.md dramatically improves the quality and consistency of Claude Code's output.

An effective CLAUDE.md includes several categories of information. First, project overview: what the project does, its architecture, and the key technologies used. Second, coding conventions: naming patterns, file organisation rules, preferred patterns and anti-patterns, and style guide references. Third, important context: environment setup requirements, external dependencies, API contracts, and database schema notes. Fourth, task-specific instructions: how to write tests in this project, how to structure new components, how to handle errors, and how to format commit messages.

Here is a practical example of what a CLAUDE.md section might look like for coding conventions: "Use TypeScript strict mode. Functional components with hooks, no class components. Tailwind CSS for styling, no inline styles. All API responses use the ApiResponse<T> wrapper type. Error handling uses the Result pattern from src/lib/result.ts. Tests use Vitest with React Testing Library."

The CLAUDE.md file is also hierarchical. You can place a CLAUDE.md at the project root for global conventions and additional CLAUDE.md files in subdirectories for module-specific instructions. Claude Code merges them contextually, applying the most relevant instructions based on which files it is working with.

For teams, CLAUDE.md serves as living documentation of your project's conventions — documentation that is not just read by humans but actively enforced by the AI. When a new developer joins the team, the CLAUDE.md gives them (and their Claude Code instance) an immediate understanding of how the project works. The [CLAUDE.md Generator tool](/tools/claude-md-generator) helps you create a comprehensive CLAUDE.md for your project based on a series of guided questions about your codebase, conventions, and workflow preferences.

Hooks and MCP: extending Claude Code's capabilities

Hooks and MCP (Model Context Protocol) are advanced features that extend Claude Code from a standalone coding agent into an integrated part of your development infrastructure.

Hooks are custom scripts that run automatically before or after specific Claude Code actions. They enable you to enforce project standards and automate quality checks as part of the AI-assisted development workflow. Common hook use cases include: running a linter after every code change to ensure style compliance, executing tests after modifications to catch regressions immediately, validating import patterns to prevent architectural violations, and triggering build checks before finalising changes.

Hooks are configured in your project settings and can be any executable script — shell scripts, Python scripts, Node.js scripts, or compiled binaries. The hook system provides context about what Claude Code just did (which files were modified, what type of change was made), allowing hooks to be targeted and efficient rather than running blanket checks on every action.

MCP (Model Context Protocol) is an open protocol that lets Claude Code connect to external tools and data sources. Through MCP servers, Claude Code can access: databases (query your development or staging database to understand schema and data), documentation systems (pull in API docs, design specs, or requirements), project management tools (read ticket details, update statuses), monitoring systems (check error logs, performance metrics), and any custom tool your team builds an MCP server for.

The practical impact of MCP is that Claude Code stops being limited to what it can see in your local filesystem. With MCP connections, it can understand the full context of a development task: the ticket requirements from Jira, the API specification from your docs platform, the current database schema, and the recent error logs from production. This broader context produces more accurate and more complete code changes.

Setting up MCP involves running MCP servers (either pre-built ones from the community or custom ones for your internal tools) and configuring Claude Code to connect to them. The MCP ecosystem is growing rapidly, with servers available for popular databases, documentation platforms, and development tools. For a comprehensive guide to MCP, including setup instructions and available servers, see our dedicated article on [What Is MCP](/blog/what-is-mcp) and the [MCP curriculum module](/coding/claude-code-mcp).

Claude Code vs Cursor vs GitHub Copilot: choosing the right tool

The three leading AI coding tools take fundamentally different approaches, and understanding these differences is essential for choosing the right one for your workflow.

**Claude Code** is an agentic, terminal-based tool. It excels at complete task execution: "implement this feature," "refactor this module," "add tests for this component." It understands your entire codebase (leveraging Claude's 1M-token context window), operates on your actual files, and can execute multi-step tasks autonomously. Its strength is handling complex, multi-file tasks that require understanding project architecture. Its limitation is that it operates outside your editor — you use it alongside your editor, not inside it.

**Cursor** is an AI-native code editor. Built as a fork of VS Code, it integrates AI deeply into the editing experience. Features include multi-file editing, codebase-aware chat, and an AI composer that can generate and modify code across files. Cursor sits between Claude Code's agentic approach and Copilot's assistive approach — it can handle multi-file tasks but operates within the editor interface. Its strength is providing a familiar editor experience with deeply integrated AI capabilities. Its limitation is that switching to a new editor is a significant decision, and its AI capabilities are tied to the editor.

**GitHub Copilot** is an assistive, editor-integrated tool. It runs as a plugin in VS Code, JetBrains, and other editors, providing inline code completions as you type. Copilot is the lightest touch — it works with your existing editor and workflow, suggesting code without changing how you develop. Its strength is seamless integration and low disruption. Its limitation is that it operates at the line or function level, not at the project level — it does not understand your full codebase architecture in the way Claude Code or Cursor does.

**How to choose.** If you work on complex, multi-file tasks and are comfortable with terminal-based workflows, Claude Code's agentic approach offers the deepest capabilities. If you want a single, integrated environment with strong AI and are willing to switch editors, Cursor provides the most cohesive experience. If you want to add AI to your existing workflow with minimal disruption, Copilot is the lowest-friction option. Many developers use combinations — Claude Code for complex tasks, Copilot for inline completions. For a detailed feature-by-feature comparison, see our [Claude Code vs Cursor analysis](/compare/claude-code-vs-cursor).

Use cases, pricing, and getting the most from Claude Code

Claude Code delivers the highest value on specific categories of development work. Understanding these helps you direct the tool where it will save the most time and produce the best results.

**High-value use cases.** Cross-codebase refactoring: renaming patterns, migrating libraries, updating API usage across hundreds of files. Test writing: generating comprehensive unit and integration tests that understand your existing testing patterns. Documentation: producing accurate, up-to-date documentation by reading the code directly. Feature implementation: building complete features from descriptions, including code, tests, and documentation. Bug investigation: analysing error reports, tracing through code to identify root causes, and implementing fixes. Code review preparation: having Claude Code review changes for potential issues before human review.

**Where Claude Code is less effective.** Highly exploratory, creative architecture work where the direction is unclear benefits from human judgment and whiteboard thinking before Claude Code involvement. Real-time pair programming during rapid prototyping may be better served by inline tools like Copilot. And any task requiring production system access or deployment should involve human oversight of every step.

**Pricing.** Claude Code usage is billed through your Anthropic account. With a Claude Pro subscription ($20/month), you get included usage. Claude Team ($25/user/month) adds collaborative features. For higher-volume usage, API-based billing charges per token — which can be more cost-effective for heavy use. The actual cost per task varies with codebase size and task complexity, but typical feature implementation tasks cost between $0.50 and $5.00 in API tokens.

**Getting the most from Claude Code.** Start with a well-written CLAUDE.md — this alone can double the quality of Claude Code's output. Always work on a Git branch so you can review and revert changes cleanly. Be specific in your instructions: "Add input validation to the POST /users endpoint following the validation pattern in POST /orders" is dramatically better than "add validation." Review Claude Code's output as you would review a junior developer's pull request — it produces high-quality code, but human oversight remains essential. And invest time in hooks and MCP integration once you are comfortable with the basics — these features multiply Claude Code's capabilities significantly. The [Coding section](/coding) of the Enigmatica curriculum covers these practices in depth.

Related Terms