Claude Code: The "Open Claw" AI CLI That Lives Inside Your Codebase

Claude Code (nicknamed "open claw" by developers for its distinctive claw-bracket icon >_) is Anthropic's agentic CLI tool that runs inside your terminal, reads your actual codebase, writes files, runs shell commands, and executes multi-step tasks autonomously. It's not a chat interface — it's an AI that operates directly on your project.

This is a complete guide covering installation, daily workflows, the CLAUDE.md configuration file, hooks, MCP server integration, and the real-world scenarios where Claude Code saves the most time.

Installation & Setup

# Prerequisites: Node.js 18+ required
node --version  # must be >= 18

# Install Claude Code globally
npm install -g @anthropic-ai/claude-code

# Verify installation
claude --version

# Authenticate (opens browser for OAuth)
claude

# Or set API key directly
export ANTHROPIC_API_KEY=sk-ant-...

After authentication, run claude from any project directory and it immediately has access to your full file structure. It will read files, understand the codebase, and ask clarifying questions before making changes.

Key Features That Make Claude Code Different

CLAUDE.md — Your Project's AI Brain

The most powerful Claude Code feature most developers underuse is the CLAUDE.md file. Place it in your project root and Claude Code reads it at the start of every session — giving it persistent context about your project without you repeating yourself.

# CLAUDE.md example

## Project: Property Management App
**Stack:** Ionic 7 + Angular 17 + Laravel 11 + MySQL

## Code Conventions
- Use TypeScript strict mode everywhere
- Services are injectable singletons in core/services/
- All API calls go through ApiService — never use HttpClient directly in components
- Use NgRx Signals for state management (not BehaviorSubjects)
- Laravel controllers use Form Requests for all validation
- Return format: { success: bool, data: any, message: string }

## Architecture Rules
- Feature modules are lazy-loaded — never import them eagerly
- Sensitive data uses @capacitor/preferences, never localStorage
- All DB queries use Eloquent — no raw SQL except for complex reports

## Testing
- Run tests: `npm run test` (Ionic) and `php artisan test` (Laravel)
- Coverage minimum: 80% for services and controllers

## DO NOT
- Use `any` type in TypeScript without a comment explaining why
- Write direct SQL queries in controllers
- Store auth tokens in localStorage

Slash Commands

Claude Code has built-in slash commands for common operations:

Command	What it does
/help	Show all available commands
/clear	Clear conversation history (start fresh)
/compact	Summarise and compress the conversation to free up context window
/model	Switch between Claude models mid-session
/review	Review a pull request (with GitHub MCP)
/commit	Auto-generate a git commit message from staged changes
/init	Analyse the project and auto-generate a CLAUDE.md file
/cost	Show token usage and API cost for the current session

Hooks — Automate Claude Code Behaviour

Hooks let you run shell commands automatically in response to Claude Code events — before/after tool calls, when files are saved, when tests run. Configured in ~/.claude/settings.json:

// ~/.claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write",
        "hooks": [
          {
            "type": "command",
            "command": "npx eslint --fix ${file}"
          }
        ]
      }
    ],
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "echo 'Running: ${command}' >> ~/.claude/audit.log"
          }
        ]
      }
    ]
  }
}

Practical hook uses: auto-run ESLint after file edits, auto-format with Prettier, log all shell commands for audit, run affected tests after each edit.

MCP Servers — Extending Claude Code's Reach

Model Context Protocol (MCP) servers give Claude Code access to external systems. Configure them in ~/.claude/settings.json:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..." }
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/projects"]
    },
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    }
  }
}

With the GitHub MCP server, Claude Code can read PRs, create issues, review code, and push branches. With the Postgres MCP server, it can introspect your actual database schema before writing migration files.

Real Workflows I Use Daily

1. New feature from ticket to commit

2. Bug fix from error log

3. Codebase-wide refactor

4. Generate tests for existing code

Permission Modes & Safety

Claude Code operates in one of three permission modes:

• Default (ask): Prompts you to approve every file write and shell command — safest for unfamiliar codebases
• Auto-approve: Approve a specific tool type for the session — e.g. always allow file writes, always ask about shell commands
• YOLO mode (--dangerously-skip-permissions): Fully autonomous — never asks, does everything. Use only in isolated/sandboxed environments