← All Cheatsheets

Claude Code

CLI commands · Slash commands · Memory · Hooks · MCP · Best practices

⚡

Getting Started

# Install

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

# Launch in project directory

claude

# One-shot command

claude -p "explain this codebase"

# Continue last session

claude --continue

Config file~/.claude/settings.json

Project memoryCLAUDE.md in repo root

AuthANTHROPIC_API_KEY env var

Slash Commands

/helpShow all available commands

/clearClear conversation context

/compactCompress context to save tokens

/memoryView and edit memory files

/initCreate CLAUDE.md for the project

/reviewReview recent changes

/costShow token usage and cost

/modelSwitch model mid-session

/doctorCheck environment health

/bugReport a bug to Anthropic

🧠

Memory System

CLAUDE.md (Project)

Place in repo root. Auto-loaded every session. Store architecture decisions, coding standards, common commands.

~/.claude/CLAUDE.md (User)

Global preferences across all projects. Preferred languages, style, tools.

# CLAUDE.md example

## Stack

- TypeScript, Next.js 14, Tailwind

## Commands

- Build: npm run build

- Test: npm test --run

## Standards

- No default exports except pages

🚀

CLI Flags

Flag	Short	Description
--print	-p	Non-interactive: print response and exit
--continue	-c	Resume the most recent conversation
--resume <id>		Resume a specific conversation by ID
--model <name>		Specify model (claude-opus-4, sonnet-4, etc.)
--output-format		json \| text \| stream-json for scripting
--max-turns <n>		Limit agentic loop iterations
--allowedTools		Comma-separated list of permitted tools
--no-tools		Disable all tools (pure chat mode)
--dangerously-skip-permissions		Skip all permission prompts (CI use)

🔌

MCP — Model Context Protocol

MCP servers extend Claude Code with custom tools — databases, APIs, internal services.

# Add MCP server

claude mcp add my-server

# List MCP servers

claude mcp list

# Config location

.claude/settings.json

Scopelocal / user / project

Transportstdio or SSE

Auto-approveautoApprove: ['tool_name']

🪝

Hooks

Automate actions triggered by Claude Code events. Defined in .claude/hooks/

PreToolUseRun before any tool call — validate, log, block

PostToolUseRun after tool call — audit, notify

StopRun when agent finishes — run tests, format

NotificationTriggered on agent notifications

# Hook runs npm test after agent stops

{"hooks": {"Stop": [{"type": "command", "command": "npm test"}]}}

🔒

Permissions & Trust

Claude Code asks permission before writing files, running commands, or accessing the network. Configure trust levels carefully.

DefaultPrompt for each sensitive action

Trust projectAuto-approve tools in this repo

Trust allAuto-approve everything (use with care)

Review diffs before approveUse --allowedToolsAudit hooksPin model version

✅

Best Practices

→

Write a good CLAUDE.md: Include stack, commands, conventions, and gotchas. Claude reads it every session.

→

Use /compact often: Compress context when working on long tasks to avoid hitting limits.

→

Be specific in tasks: Give Claude a clear goal, acceptance criteria, and constraints upfront.

→

Commit frequently: Commit before and after Claude makes changes so you can easily revert.

→

Use headless mode for CI: claude -p "run tests and fix failures" --dangerously-skip-permissions

→

Chain with pipes: cat error.log | claude -p "explain this error"

→

Use --output-format json: Parse Claude output programmatically in scripts.

→

Custom slash commands: Add .claude/commands/my-cmd.md for project-specific shortcuts.