My Claude Code Setup

My Claude Code Setup

A complete walkthrough of how I configure Claude Code for maximum productivity.

What is Claude Code?

Claude Code is Anthropic's official CLI tool that brings Claude directly into your terminal. It's not just a chatbot—it can read files, execute commands, edit code, and integrate with external tools via MCP (Model Context Protocol).

Installation

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

Or if you prefer:

brew install claude-code

Configuration Files

Claude Code uses several configuration files in ~/.claude/:

settings.json

This is the main configuration file. Here's what matters:

{
  "model": "opus",
  "alwaysThinkingEnabled": true,
  "max_turns": 100,
  "includeCoAuthoredBy": false
}

Key settings:

• model: I use opus for complex tasks (expensive but reliable), sonnet for everyday work
• alwaysThinkingEnabled: Without thinking even Opus is weak
• max_turns: How many back-and-forth turns before stopping
• includeCoAuthoredBy: Avoid having "Co-authored-by" added to git commits

Permissions

Claude Code has a robust permission system. I prefer a permissive setup with explicit allows:

{
  "permissions": {
    "allow": [
      "*",
      "Bash(go build:*)",
      "Bash(go test:*)",
      "Bash(git commit:*)",
      "Bash(make:*)"
    ],
    "defaultMode": "default"
  }
}

The * allows most operations, but you can be more restrictive.

Usually I run Claude with:

claude --dangerously-skip-permissions --resume --chrome

I know, it's dangerous, but I commit continuously and have good backup strategies. This will speed up your work by 10x.

Resume is usefull for a list of previous esession to continue and --chrome will allow browser acces to understand the web content you're building (you need the Claude extension).

CLAUDE.md

This file lives in ~/.claude/CLAUDE.md (global) or in your project root (project-specific). It's your way to give Claude persistent instructions.

My global CLAUDE.md:

# Style
Apply KISS, YAGNI, DRY and SOLID coding principles.
Always respond in English.

# Rules
- Do not duplicate logic
- Do NOT stop until all tasks are complete
- Check your todo list after each step
- Never skip failing tests. Fix them!

MCP Servers

MCP (Model Context Protocol) extends Claude's capabilities. I use three:

mcp.json

{
  "mcpServers": {
    "serena": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "git+https://github.com/oraios/serena", "serena-mcp-server"]
    },
    "chrome-devtools": {
      "type": "stdio",
      "command": "npx",
      "args": ["chrome-devtools-mcp@latest"]
    },
    "context7": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@upai/mcp-server-context7"]
    }
  }
}

What each does:

• Serena: Code intelligence—symbol search, references, project navigation
• Chrome DevTools: Browser automation and debugging
• Context7: Up-to-date documentation lookup for any library

Plugins

Claude Code has a plugin system. My enabled plugins:

Enable in settings:

{
  "enabledPlugins": {
    "context7@claude-plugins-official": true,
    "feature-dev@claude-plugins-official": true,
    "claude-hud@claude-hud": true
  }
}

Hooks

Hooks let you run commands before/after Claude actions. I use them for automatic git-ai checkpoints:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit|MultiEdit",
        "hooks": [
          {
            "type": "command",
            "command": "git-ai checkpoint claude --hook-input stdin"
          }
        ]
      }
    ]
  }
}

This creates automatic checkpoints to distinguish when AI edited the code and when it was a human activity. Thanks to https://github.com/acunniffe/git-ai

Status Line

The status line shows useful info while Claude works. I use claude-hud(https://github.com/jarrodwatts/claude-hud):

{
  "statusLine": {
    "type": "command",
    "command": "bash -c 'node \"$(ls -td ~/.claude/plugins/cache/claude-hud/claude-hud/*/ | head -1)dist/index.js\"'"
  }
}

iTerm 2 Notifications

Never miss when Claude completes a task. In iTerm 2:

1. Open Preferences → Profiles → Terminal
2. Enable "Silence bell"
3. Select "Send escape sequence-generated alerts"
4. Set your preferred notification delay

This sends a system notification when Claude finishes long-running tasks.

Makefile Patterns

Every project gets a Makefile. It's the universal interface that Claude understands immediately.

Key patterns:

• Self-documenting help: make help parses ## target: description comments
• Consistent targets: build, test, deploy, clean work the same everywhere
• Environment loading: Auto-include .env files when present

The value is consistency. Claude reads make deploy and knows what to do—no explaining build systems per project.

Tips for Daily Use

1. Almost always use Plan mode - Let Claude think before acting
2. Give Claude verification methods - Unit tests, Chrome extension, or simulators
3. Hold the same bar for human and Claude code - Use /code-review to automate review
4. Press Shift+Tab - Toggle auto-accept for file edits, reduces friction
5. Use opus for complex tasks - Worth the cost for architecture decisions
6. Check the todo list - Claude tracks its own progress, review it
7. Leverage MCP - Context7 for docs, Serena for code navigation
8. Trust but verify - Always review generated code before committing

Project-Specific Config

Each project can have its own .claude/CLAUDE.md with specific instructions:

# Project: MyApp

## Stack
- Go 1.22
- PostgreSQL 15
- React 18

## Commands
- `make test` - Run tests
- `make build` - Build binary

## Conventions
- Use hexagonal architecture
- All errors must be wrapped with context

Conclusion

Claude Code is more than a coding assistant—it's a development environment that understands context, follows instructions, and integrates with your tools. The key is configuration: take time to set up your CLAUDE.md, permissions, and MCP servers.

The investment pays off exponentially.