Building AI Agents with Claude Agent SDK

This hands-on course introduces you to the Claude Agent SDK, Anthropic's official Python SDK for building AI-powered agents that can reason, use tools, and automate real-world tasks.

You will learn how to create agents that interact with cloud infrastructure, use file operations, manage multi-turn conversations, and orchestrate complex workflows. Through progressive labs, you will build agents from simple single-turn queries to multi-agent systems that provision and manage AWS infrastructure declaratively, combining Skills, Subagents, and Model Context Protocol (MCP) integrations.

Course Highlights

Getting Started with the Claude Agent SDK

This module takes you from understanding the agent loop to building context-aware, multi-turn agents. You will learn how the Claude Agent SDK enables Python developers to create agents that think, act, observe, and decide autonomously.

Through progressive labs, you will execute your first query, master the SDK's built-in tools for file and infrastructure operations, and build multi-turn conversational agents that iteratively develop AWS CloudFormation templates.

Introduction to the Claude Agent SDK

• Install the Claude Agent SDK and verify the environment setup
• Understand the agent loop pattern: Think, Action, Observe, Decide
• Configure ClaudeAgentOptions with model, tools, and permission modes
• Execute your first agent query and extract responses from AssistantMessage objects
• Use the Bash tool to interact with AWS CLI for identity verification

Mastering Claude Tools

• Use discovery tools such as Glob for file pattern matching, Read for file contents, and Grep for content search
• Apply modification tools such as Edit for targeted changes, MultiEdit for batch operations, and Write for file creation
• Execute shell commands with the Bash tool for Terraform deployment workflows
• Debug and fix broken Terraform infrastructure using agent-driven file operations

Multi-Turn Conversations and CloudFormation

• Initialize a ClaudeSDKClient for persistent, multi-turn conversations
• Preserve conversation context across multiple exchanges using a single client instance
• Configure permission modes for safe file editing and infrastructure provisioning
• Iteratively develop and enhance AWS CloudFormation templates through progressive conversation turns

Skills and EC2 Operations

• Create reusable Skills as filesystem-based artifacts with YAML frontmatter configuration
• Structure skill files in the .claude/skills/ directory with the appropriate SKILL.md format
• Enable skill loading in agent configuration with setting sources and tool permissions
• Build an EC2 management skill for instance operations using AWS CLI
• Orchestrate multiple skills together for complex infrastructure workflows

Subagents and VPC Automation

• Define specialized Subagents programmatically using the AgentDefinition class
• Configure automatic task delegation based on subagent descriptions
• Apply the principle of least privilege by restricting tools per subagent
• Build a network agent and a security agent with clearly separated concerns
• Orchestrate multi-turn VPC provisioning with coordinated subagent collaboration

MCP and AWS CloudFormation

• Connect to MCP servers as external tool providers using stdio transport
• Discover and list available MCP tools with namespaced naming conventions
• Use MCP tools in agent conversations for CloudFormation operations
• Build complete MCP-driven workflows for template validation and stack deployment
• Coordinate multiple MCP servers for filesystem and CloudFormation management