Claude Code — CLI Setup, MCP & Daily Usage Guide

claude-code-setup.md

Claude Code — Setup & Reference

Install

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

Requires Node.js 18+.

---

Run

claude                          # interactive session
claude "your prompt" --print    # one-shot, print and exit

Flag	Purpose
--model claude-opus-4-6	Override model
--print	Non-interactive output
--no-stream	Disable streaming

---

GitHub MCP Setup

Step 1 — Create a Fine-Grained PAT

GitHub → Settings → Developer settings → Personal access tokens → Fine-grained tokens

Permission	Level
Contents	Read and Write
Issues	Read and Write
Pull requests	Read and Write
Metadata	Read (auto)

Step 2 — Add MCP Servers

Run once in terminal. Replace YOUR_TOKEN with your PAT.

claude mcp add github -s user -e GITHUB_PERSONAL_ACCESS_TOKEN=YOUR_TOKEN -- npx -y @modelcontextprotocol/server-github
claude mcp add filesystem -s user -- npx -y @modelcontextprotocol/server-filesystem /your/work/directory
claude mcp add fetch -s user -- npx -y @modelcontextprotocol/server-fetch

Restart Claude Code after adding servers.

Step 3 — Verify

/mcp

Lists all connected servers and tools. Ask Claude "list my GitHub repos" to confirm it's working.

---

Quick Reference

Slash Commands

Command	What it does	When
/compact	Compress context, keep summary	Before starting a new task in a long session
/clear	Full context wipe	Switching to a completely different task
/cost	Show token usage	Check if context is getting full
/mcp	List connected MCP servers	Verify tools are loaded
/model	Show or switch model	Use faster model for simple tasks
/memory	View/edit memory files	Review what Claude has saved
/help	All commands	Reference

Other MCP Servers

Package	Purpose
@modelcontextprotocol/server-postgres	Query PostgreSQL
@modelcontextprotocol/server-brave-search	Web search
@modelcontextprotocol/server-slack	Slack messages

Browse all: https://github.com/modelcontextprotocol/servers

---

Config Files

File	Purpose
~/.claude.json	MCP servers, API key, global defaults
~/.claude/settings.json	Auto-approve, theme, model
~/.claude/skills/	Custom slash commands
CLAUDE.md (project root)	Project instructions — loaded every session

settings.json example

{
  "theme": "dark",
  "model": "claude-sonnet-4-6",
  "autoApprove": ["read", "glob", "grep", "bash(git *)"]
}

---

CLAUDE.md

Put a CLAUDE.md in your project root. Claude reads it at session start every time — use it for rules, stack info, and anything you'd otherwise repeat every session.

---

Memory System

Claude auto-saves context across sessions at ~/.claude/projects/<hash>/memory/.

Type	Stores
user	Your role, preferences, background
feedback	Corrections and confirmed approaches
project	Ongoing work, decisions, deadlines
reference	Where to find things (dashboards, trackers)

---

Skills (Custom Commands)

Create ~/.claude/skills/my-skill.md:

---
name: my-skill
description: What it does
---

Steps for Claude to follow when invoked.

Invoke with /my-skill or /my-skill some args.

---

Effective Usage

• Be specific upfront — "update the README, remove X, add Y" beats "update the readme" every time
• Number your tasks — list 5 things as 1–5, not a paragraph; Claude tracks them in order
• Correct immediately — don't let a wrong direction run for 3 messages
• Use /compact proactively — after finishing a big task, before starting the next one
• Multi-task in one message — Claude runs independent tasks in parallel, saves round trips
• Use CLAUDE.md for rules — if you say the same thing every session, put it there instead

---

Keyboard Shortcuts

Shortcut	Action
Esc	Cancel current operation
Ctrl+C	Exit
↑ / ↓	Navigate history
Shift+Enter	New line in prompt