Intermediate

What is Claude Code

Claude Code is an agentic command-line tool that gives Claude direct access to your local development environment. Unlike the Claude.ai web interface, Claude Code can read your files, write code, run shell commands, and iterate autonomously on multi-step tasks — all from your terminal, inside your actual project.

What Makes Claude Code Different from Claude.ai Chat

Claude.ai (web chat)

You paste code snippets manually
Claude reads only what you share in the conversation
Claude generates code but does not run it (except in React Artifact sandbox)
No access to your file system, terminal, or tools
Each conversation starts with no project context

Claude Code (CLI)

Claude reads your actual project files directly
Claude can explore the full codebase — you don't need to paste anything
Claude can run shell commands, tests, and build scripts
Claude can write files and make changes directly
CLAUDE.md provides persistent project context across sessions

What Claude Code Can Do

Claude Code has access to a set of tools that give it agency over your local environment:

Read files: Explores your project structure, reads any file you have access to
Write files: Creates new files and edits existing ones — with your approval (unless pre-approved)
Run shell commands: Execute test suites, build scripts, linters, git commands, and arbitrary terminal commands
Search codebase: Grep, glob, and search across files to find relevant code without you specifying exact paths
Browse the web: Fetch documentation, check API references, or look up current information (when enabled)
Call MCP tools: Access external tools and services via the Model Context Protocol (databases, APIs, cloud services)

How Claude Code Sessions Work

You start a Claude Code session by running claude in your project directory. Claude Code then:

Reads any CLAUDE.md files in the project hierarchy (project instructions + memory)
Enters an interactive REPL — you type tasks, Claude executes them
Prompts you for permission before file writes and shell commands (unless pre-approved)
Maintains conversation context within the session — earlier results inform later actions
Saves memory automatically for the next session (if configured)

Sessions are stateful within a single invocation. Context does not persist between separate claude invocations by default — CLAUDE.md is the mechanism for cross-session memory.

When to Use Claude Code vs the Web Interface

Use Claude Code for: multi-file code changes, debugging with actual error output, running tests and iterating on failures, refactoring a codebase, understanding an unfamiliar project, build and deployment automation
Use Claude.ai for: one-off questions, writing and editing tasks not related to code, research, conversation-oriented tasks, quick code snippets you just want to copy
Use both together: Claude.ai for planning and research; Claude Code for implementation — they share the same underlying model but have different tool access

Claude Code's Approach to Safety

Claude Code is designed around the principle of minimal footprint — it asks before taking actions with side effects, prefers reversible actions over irreversible ones, and stops to ask when uncertain rather than proceeding autonomously.

Defaults are cautious: file writes and shell commands require explicit approval unless you pre-approve them in settings. You can widen permissions to streamline your workflow, but the conservative defaults exist to prevent unintended changes to your codebase.

Checklist: Do You Understand This?

Claude Code is a CLI with direct file system and shell access — not a chat interface wrapper
Key capabilities: read files, write code, run shell commands, search codebase, call MCP tools
Sessions start fresh each invocation — CLAUDE.md provides the cross-session project context
Use Claude Code for multi-step code tasks; use Claude.ai for writing, research, and quick questions
Default permissions are cautious — Claude asks before writes and shell commands unless pre-approved