What you'll learn

Chatbots are easy to prototype, but they show limitations when the job requires more than a single response. Once your application needs to inspect files, call tools, coordinate subtasks, enforce permissions, recover from failures, and show what happened, then you need more than a prompt, you need an agent workflow. In this course, Building with Claude Agent SDK, you’ll learn how to create reliable Claude agents in Python. First, you’ll explore where the Claude Agent SDK fits among direct API usage, Claude Code, and Managed Agents, and how SDK agents run through sessions, tools, permissions, and streamed messages. Next, you’ll configure your first agent, load project context, manage memory and session options, and inspect the agent’s tool calls and results programmatically. Then, you’ll build multi-agent workflows with specialized subagents, add custom MCP tools and external service integrations, and govern tool use with permissions and hooks. Finally, you’ll test, debug, evaluate, monitor, and deploy agent workflows safely. When you’re finished, you’ll be able to use the Claude Agent SDK to build Python agents that can complete real work while remaining observable, extensible, and safe.