Claude Code Tutorial and Workflows: Agentic CLI Setup, MCP Tools & Architecture Guide

A definitive technical guide to setting up Anthropic’s Claude Code CLI, configuring Model Context Protocol (MCP) servers, executing multi-turn autonomous bash commands, and building automated developer workflows. As coding agents move from graphical desktop IDEs directly into the Unix terminal, Anthropic’s Claude Code introduces an agentic command-line interface designed to edit repositories, execute shell commands, run test suites, and manage git branches autonomously.

Claude Code Architecture: How It Operates in Your Shell

Claude Code is built on top of Anthropic’s flagship Claude 3.5 Sonnet and Claude 3.7 Sonnet reasoning engines. Rather than uploading entire repository files over the wire, Claude Code maintains an active tool-calling loop executed directly within your local shell environment:

• File System Explorer: Uses high-performance local binaries (like ripgrep and fd) to search file names and content patterns without bloating context window tokens.
• Atomic File Patching: Employs structured line-number target replacements, ensuring formatting, comments, and untouched code sections remain strictly intact.
• Command Execution Sandbox: Spawns non-interactive subshells to run linters, test harnesses, package managers, and git commands, feeding stdout and stderr back into the reasoning loop.
• Model Context Protocol (MCP) Integration: Natively connects to external databases, Docker daemons, GitHub APIs, and web search servers through standardized MCP JSON-RPC protocols.

Step-by-Step Installation and Initial Setup

Claude Code requires Node.js (version 18 or higher) and an active Anthropic Console account or Anthropic API key.

1. Installing the Global CLI Binary

Install Claude Code globally via npm or homebrew:

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

# Verify installation and view version
claude --version

2. Authenticating Your Account

Launch Claude Code inside any local project repository to begin the OAuth web authentication flow:

cd ~/projects/my-app
claude

On first launch, Claude Code opens your default browser to authorize your Anthropic Console account. Once authenticated, your authentication token is stored securely in ~/.claude/config.json.

Essential Claude Code Commands and Shortcuts

Once inside the interactive Claude Code CLI prompt, use these built-in slash commands to manage sessions and context:

The Power of CLAUDE.md: Structuring Repo Instructions

Just as Cursor uses .cursorrules, Claude Code relies on a dedicated markdown guide placed in your repository root named CLAUDE.md. When Claude Code initializes, it reads this file automatically before processing any prompt.

A well-architected CLAUDE.md should define:

# Project Guidelines for Claude Code

## Build & Test Commands
- Run development server: `npm run dev`
- Run test suite: `npm test`
- Run single test: `npm test -- path/to/test.ts`
- Lint & format check: `npm run lint`

## Architecture Conventions
- All database queries must route through Prisma client in `lib/prisma.ts`.
- Never use inline `any` types in TypeScript; use strict generic interfaces.
- Component styling relies strictly on Tailwind CSS classes; do not create `.css` modules.
- Ensure all public API routes validate incoming payloads with Zod schemas.

Configuring Model Context Protocol (MCP) Servers

Claude Code’s real superpower is its native support for the Model Context Protocol (MCP). By connecting MCP servers, you grant Claude direct access to PostgreSQL databases, remote GitHub repositories, and AWS infrastructure.

To register an MCP server with Claude Code, run:

# Connect a local PostgreSQL database via MCP
claude mcp add postgres npx -y @modelcontextprotocol/server-postgres postgresql://user:password@localhost:5432/mydb

# Connect GitHub MCP for PR and issue automation
claude mcp add github npx -y @modelcontextprotocol/server-github

Once connected, Claude Code can inspect live database schemas, verify foreign key constraints, and query actual records to troubleshoot database bugs without leaving your terminal.

Real-World Production Workflows

Workflow 1: Autonomous Test-Driven Bug Fixing

When an automated test fails on CI, pipe the error trace directly to Claude Code:

claude "Our checkout test is failing with error: OrderTotalMismatch. Locate the calculation logic, write a reproduction unit test, fix the bug, and run npm test to verify."

Claude Code reads the failing test file, discovers the calculation discrepancy in taxService.ts, applies the correct mathematical rounding rule, runs npm test, and reports a clean green pass.

Workflow 2: Automated Branch Creation & Pull Request Lifecycle

Automate end-to-end feature development from terminal to GitHub:

claude "Create a new branch feature/webhook-retry. Implement exponential backoff for failed webhook deliveries using BullMQ, commit the changes with conventional commit syntax, and open a PR against main."

Token Economics and Cost Optimization Tips

Because Claude Code operates continuously in an agentic loop, optimizing token consumption is crucial:

• Keep Context Compact: Regularly run /compact every 10 to 15 turns to summarize redundant file reads and output logs.
• Use Specific Prompts: Rather than saying “Refactor billing,” say “Refactor src/services/billing.ts to use async/await and update its Jest tests in tests/billing.test.ts.”
• Leverage Ignore Files: Claude Code respects your .gitignore file. Ensure large build directories (node_modules/, dist/, .next/, build/) are properly ignored to prevent accidental context pollution.