Claude Code FAQ

Short answers, no filler. Typical spend is around $6 a day. Cursor and Claude Code solve different problems. Skills are reusable expertise folders. Offline mode does not exist. Everything below is a question builders ask before they commit.

Is Claude Code free?

Installation is free. Running it is not. You need either an Anthropic API key or a Claude subscription. API billing is per token, and daily spend lands near $6 for typical development work, with 90% of users under $12 a day. Prefer flat pricing? Claude Pro at $20/month bundles Claude Code with usage caps. Max plans run $100-200/month for higher ceilings and Claude 4 Opus access.

# Native installer (recommended) - no Node.js required
curl -fsSL https://claude.ai/install.sh | bash  # macOS/Linux
# Windows: irm https://claude.ai/install.ps1 | iex
 
claude  # Prompts for API key on first run

The native installer guide covers one-command setup. The full installation guide covers every other option.

What models does Claude Code use?

The Claude 4 family powers everything. Claude 4 Sonnet is the daily default: fast, capable, cheap enough to run all day. Claude 4 Opus thinks harder on architectural calls but slows down and costs more.

Switching is a short word away. The aliases are sonnet, opus, haiku, and opusplan. That last one is a hybrid: Opus handles the plan, Sonnet handles the execution. Strong reasoning where it matters, lower bills everywhere else.

# Use aliases for quick switching
claude --model sonnet
/model opus
 
# Hybrid mode: Opus plans, Sonnet executes
/model opusplan
 
# Set a permanent default in settings.json
# { "model": "claude-sonnet-4-20250514" }

Model priority, highest to lowest: /model during a session > --model at startup > ANTHROPIC_MODEL env var > model in settings.json. Sub-agents get their own knob, CLAUDE_CODE_SUBAGENT_MODEL, if you want them on a different tier.

Claude Code vs Cursor: what's the difference?

Two tools, two jobs. Claude Code lives in the terminal as an AI agent. It reads your whole codebase, runs commands, and handles multi-file changes on its own. Cursor is an AI-first IDE built around real-time completion and inline suggestions. The one-liner: Claude Code = AI drives, you supervise. Cursor = you drive, AI assists. Plenty of developers run both at once, Cursor for editing and Claude Code for deeper reasoning and automation.

What are Claude Code Skills?

Skills are expertise you package once and reuse forever. Each Skill is a folder holding instructions, scripts, and resources, and Claude auto-loads the relevant one when context calls for it. A code-review skill, for example, hands Claude your team's standards every time without you typing them out. Personal Skills live in ~/.claude/skills/. Project Skills live in .claude/skills/.

# Skills auto-load when relevant, or invoke manually
/skill security-review

What's CLAUDE.md and why do I need it?

Chat messages fade at the end of a session. CLAUDE.md does not. It's a memory file in your project root that loads automatically on every run, feeding Claude your stack details, conventions, code style, and the commands you use most. Treat it like a strict set of rules: Claude follows CLAUDE.md more tightly than a mid-session prompt. Keep it under 300 lines.

# Auto-generate a starter CLAUDE.md
/init

How do I manage API costs?

Three habits cut the bill. Track your actual usage with ccusage. Default to Claude 4 Sonnet for routine work and save Opus for the hard stuff. Turn on prompt caching whenever you repeat operations. The /compact command squeezes context down when a session gets long. Running heavy? A Max subscription at $100-200/month usually beats pay-per-token.

# Check your usage
npx ccusage
 
# Compress context to save tokens
/compact

Can Claude Code write tests and deploy apps?

Tests, yes. Claude writes unit, integration, and end-to-end suites across Jest, Pytest, Vitest, and friends. Deployment is halfway: Claude generates scripts, Dockerfiles, and CI/CD configs, but it won't push production for you. Everything gets prepared. You type the last command yourself.

Does Claude Code work offline?

No. Every AI call routes to Anthropic's API, so the network has to be up. Your code stays on your machine. Your prompts do not. Plan for connectivity before leaning on Claude Code in a production workflow.

How do I configure Claude Code for my project?

Two systems, running side by side. CLAUDE.md files tell Claude what to do. settings.json files tell it how to operate: permissions, tools, behavior.

For the "what to do" side, CLAUDE.md in your project root carries the codebase context. Skills under .claude/skills/ package reusable workflows. Slash commands under .claude/commands/ bundle custom actions. /init writes a starter CLAUDE.md if you want one generated.

For the "how to operate" side, settings.json runs a 4-scope hierarchy: Managed (IT-deployed, top priority), Local (.claude/settings.local.json, personal to this repo), Project (.claude/settings.json, shared with your team), and User (~/.claude/settings.json, your global defaults). Narrower scopes override broader ones, so team standards go in Project and your private tweaks go in Local.

# Initialize CLAUDE.md
/init
 
# Add $schema for autocomplete in your editor
# Then edit .claude/settings.json

What can Claude Code actually do?

A lot. It builds full apps from scratch, refactors across hundreds of files, debugs by reading logs and tracing code paths, writes full test suites, drives git, and runs terminal commands. The 200K token context window means it remembers your whole conversation and the structure of the codebase you are working in. Unlike a chatbot that forgets the previous turn, Claude Code holds the full project in view.

How do I fix "command not found" errors?

Usually PATH. The installer didn't wire claude into your shell. The native installer handles this on its own, so reinstalling fixes it most of the time.

# Recommended: use native installer
curl -fsSL https://claude.ai/install.sh | bash  # macOS/Linux
# Windows PowerShell: irm https://claude.ai/install.ps1 | iex
 
# Restart terminal, then verify:
claude --version

Do I need programming experience?

Not strictly. Claude Code narrates what it's doing, suggests the next step, and teaches as it goes. Non-technical founders have shipped full MVPs using plain English. Basic terminal comfort still helps. Knowing how to cd around and run a command is the floor.

How does Claude Code handle privacy?

Every prompt goes to Anthropic's API for processing. That's the nature of the tool. Anthropic states they don't train models on API data. For sensitive codebases, the security documentation is the right starting point, and enterprise agreements can add contractual guarantees on top. Your code stays on your machine. Only what you actively share ends up in the API call.