Getting Started

Install & Launch

npm install -g @anthropic-ai/claude-code
claude            # start interactive session
claude --version  # check version
claude update     # update to latest

Authentication

claude login              # browser OAuth
export ANTHROPIC_API_KEY="sk-ant-..."
claude logout             # clear session

Slash Commands

Session Commands

/help	Show available commands
/clear	Clear conversation history
/compact	Condense context to save tokens
/cost	Show token usage and cost
/status	Show session info and model
/new	Start a new conversation
/config	Open or view configuration
/doctor	Diagnose configuration issues
/init	Create a CLAUDE.md for the project
/login	Authenticate with Anthropic
/logout	Clear authentication

Workflow Commands

/bug	File a bug report
/review	Request a code review
/pr	Create or update a pull request
/commit	Commit staged changes with a message

Keyboard Shortcuts

Ctrl+C	Cancel current generation
Ctrl+D	Exit Claude Code (EOF)
Escape	Cancel current input / edit
Tab	Autocomplete file paths and commands
Up/Down	Navigate input history

CLI Flags

Common Flags

-p, --print	Non-interactive (headless) mode
--model	Set model: `opus`, `sonnet`, `haiku`
--output-format	Output as `text`, `json`, `stream-json`
--allowedTools	Restrict tools: `Edit,Read,Bash`
--max-turns N	Limit conversation turns
--debug	Enable debug logging
-n, --name	Name the session

Headless / Scripting

claude -p "explain this error" < log.txt
claude -p "list todos" --output-format json
echo "fix typos" | claude -p

Configuration Files

CLAUDE.md Hierarchy

./CLAUDE.md	Project-level instructions (checked into repo)
./.claude/settings.json	Project settings (permissions, hooks)
~/.claude/CLAUDE.md	User-global instructions (all projects)
~/.claude/settings.json	User-global settings
~/.claude/projects/*/CLAUDE.md	Per-project user instructions (not in repo)

Environment Variables

ANTHROPIC_API_KEY	API key for direct authentication
CLAUDE_MODEL	Default model override
CLAUDE_CONFIG_DIR	Custom config directory path

Permissions

settings.json

{
  "permissions": {
    "allow": ["Read", "Glob", "Grep"],
    "deny": ["Bash(rm *)"]
  }
}

Permission Modes

default	Ask before risky tools
--dangerously-skip-permissions	Allow all tools (CI/scripts only)
--allowedTools	CLI flag to restrict tool set

MCP Servers

What Are MCP Servers?

Model Context Protocol servers extend Claude Code with custom tools — databases, APIs, services. Managed via CLI or `.mcp.json`.

CLI Management

claude mcp add server-name -- cmd arg1 arg2
claude mcp list
claude mcp remove server-name

.mcp.json Config

{
  "mcpServers": {
    "my-db": {
      "command": "python",
      "args": ["-m", "mcp_server_db"],
      "env": { "DB_URL": "${DATABASE_URL}" }
    }
  }
}

Scope

.mcp.json	Project-level MCP servers
~/.claude/mcp.json	User-global MCP servers
claude mcp add -s user	Add to user scope
claude mcp add -s project	Add to project scope

Hooks

Hook Events

PreToolUse	Runs before a tool is executed
PostToolUse	Runs after a tool completes
Notification	Runs when Claude sends a notification
Stop	Runs when Claude finishes a response

settings.json Example

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Bash",
      "hooks": [{
        "type": "command",
        "command": "check-command.sh $INPUT"
      }]
    }]
  }
}

Hook Behavior

exit 0	Allow the tool to proceed
exit 2	Block the tool from running
stdout	JSON feedback to Claude

SDK & Automation

Headless Scripting

# Single prompt, get JSON output
claude -p "summarize main.py" \
  --output-format json

# Pipe input
git diff | claude -p "review this diff"

CI / GitHub Actions

- uses: anthropics/claude-code-action@v1
  with:
    claude_args: >
      --max-turns 5
      --model claude-sonnet-4-6

Tips

Use `--max-turns` to cap cost in automation. Use `--output-format json` to parse results programmatically. Combine with `--allowedTools` for safety.

Models

Model Selection

claude --model opus     # most capable
claude --model sonnet   # balanced (default)
claude --model haiku    # fastest, cheapest

Model Shortcuts

opus	Claude Opus — highest capability
sonnet	Claude Sonnet — default, balanced
haiku	Claude Haiku — fast and economical
--model full-name	e.g. `claude-sonnet-4-6`

Best Practices

Writing CLAUDE.md

Put project conventions, tech stack, build commands, and test instructions in CLAUDE.md. Keep it concise — Claude reads it every session. Use `/init` to scaffold one.

Cost Management

/cost	Check running token usage
/compact	Compress context when it grows large
--max-turns	Cap turns in automation
sonnet / haiku	Use cheaper models for simple tasks

Effective Prompting

Be specific	"Fix the null check in auth.ts:42" > "fix bug"
Give context	Reference files, functions, error messages
Use @file	Reference files directly: `@src/main.ts`
Use images	Drag & drop screenshots for visual context
Iterate	Follow up to refine; use `/clear` to reset

CLAUDE CODE QUICK REFERENCE

CLAUDE CODE QUICK REFERENCE (continued)