Claude Code Cheatsheet: Commands, Slash Commands, CLAUDE.md and MCP

Print it, pin it, or keep it in a second tab. This is the dense reference for Claude Code, Anthropic's terminal-based coding agent: the commands you will use daily, the slash commands worth memorizing, a production-ready CLAUDE.md template, and MCP wiring. For the guided walkthrough, read our Claude Code mastery tutorial afterwards.

Install and start

Essential slash commands (inside a session)

Keyboard and prompt tricks

• Esc — stop Claude mid-action. Esc twice — edit a previous message.
• Shift+Tab — cycle permission modes, including Plan mode (proposes, does not execute) and auto-accept edits.
• @path/to/file — reference a file so Claude reads it directly.
• # your note at the start of a prompt — save the note to project memory (CLAUDE.md).
• ! prefix — run a shell command yourself inside the session (for example !npm test) and keep its output in context.
• "think" / "think hard" / "ultrathink" in a prompt — ask for progressively deeper reasoning on hard problems (and accept the extra cost).
• Ctrl+R — search prompt history. Ctrl+C — interrupt/exit.

CLAUDE.md template (copy, adapt, commit it)

# Project: [Name]

## What this is
One paragraph: what the project does, who uses it, the stack.

## Commands
- Install: `npm install`
- Dev: `npm run dev`
- Test: `npm test` (single file: `npm test -- path/to/file.test.ts`)
- Lint/typecheck: `npm run lint && npm run typecheck`
- Build: `npm run build`

## Architecture
- Entry point: src/main.ts — keep it thin.
- Business logic lives in src/services/, never in route handlers.
- Data access only through src/db/ (no raw SQL in services).

## Rules
- TypeScript strict. No `any` without a comment explaining why.
- All new endpoints need a test. Run the full test suite before saying "done".
- Never edit migrations that are already merged; create a new one.
- Match the style of the file you are editing; do not reformat unrelated code.

## Gotchas
- The payments webhook is fragile: read docs/payments.md before touching it.
- Local .env uses staging keys. Never print secrets in output or logs.

Keep it short — Claude reads this every session, so every line costs context. Put long runbooks in separate docs and link them.

Connect MCP servers

# Project-scoped (shared via .mcp.json in the repo)
claude mcp add github -- npx -y @modelcontextprotocol/server-github

# From a JSON config (remote server with headers)
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

# Inspect and remove
claude mcp list
claude mcp remove github

Equivalent .mcp.json at your repo root:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/project"]
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres",
               "postgresql://user:pass@localhost:5432/app"]
    }
  }
}

New to MCP itself? Start with our MCP guide and the companion MCP servers cheatsheet.

Daily workflow that works

1. Plan first: for anything non-trivial, use Plan mode (Shift+Tab) and approve the approach before code changes.
2. Small loops: one task, run tests, review the diff, commit. Then /clear before the next unrelated task.
3. Show, do not tell: paste the error, the log, the failing test — do not describe them from memory.
4. Verify like a skeptic: ask Claude to run the test suite and the linter, and read the diff yourself. "It should work" is not verification.
5. Feed the memory: when Claude makes the same mistake twice, add one line to CLAUDE.md.

Troubleshooting quick hits

• It keeps asking permission for everything: tune /permissions or allowlist safe commands in .claude/settings.json.
• Answers feel generic: your CLAUDE.md is thin or missing; run /init and add commands and gotchas.
• Context fills instantly: stop pasting whole repos — reference files with @, use /compact, and split work across subagents.
• MCP server will not connect: claude mcp list for status, test the server's command manually in a terminal, and check that required API keys are in your environment.

More references in our cheatsheets library.