Module 1 · Harnesses · Lesson 2

The Claude Agent SDK

How query() drives Claude Code's agent loop from your own TypeScript, and how custom tools, permissions, hooks, sessions and structured output work, with the Claude Code equivalent of each.

70 min 1 video 3 exercises 7-question quiz Answers

Setup for the exercises:Install Claude Code Node.js, pnpm and TypeScript MCP servers in Claude Code Get an OpenRouter key Local models with Ollama OpenRouter or a local model?

By the end of this lesson you can

Explain what query() does, which messages it yields, and how a run ends (success, turn limit, budget limit)
Implement a custom tool with tool() and createSdkMcpServer(), and predict its model-facing name
Trace a tool call through the six-step permission evaluation order and predict whether it runs, is denied or reaches canUseTool
Use PreToolUse and PostToolUse hooks to enforce a write boundary and keep an audit trail
Map every SDK option used in this lesson to its Claude Code equivalent so you can prototype on a subscription

Claude Code is an agent you use. The Claude Agent SDK is the same agent as a library you embed. When you call query(), the SDK starts the Claude Code binary as a subprocess, hands it your prompt and options, and streams back every message the loop produces: the model's text, each tool call, each result, and a final summary with cost and a session ID. Everything Claude Code knows how to do (read and edit files, run commands, compact long conversations, load skills, spawn subagents) is available, and you decide which parts to switch on.

This lesson builds a small title-review agent with the SDK and, alongside each feature, shows the Claude Code configuration that does the same thing. SDK code needs an API endpoint, not your subscription: your own OpenRouter key (any model) or a free local model through Ollama. The Claude Code version runs on your subscription. The concepts are identical because the harness is identical.

Module 1 · Harnesses · Lesson 2

The Claude Agent SDK

How query() drives Claude Code's agent loop from your own TypeScript, and how custom tools, permissions, hooks, sessions and structured output work, with the Claude Code equivalent of each.

70 min 1 video 3 exercises 7-question quiz Answers

Setup for the exercises:Install Claude Code Node.js, pnpm and TypeScript MCP servers in Claude Code Get an OpenRouter key Local models with Ollama OpenRouter or a local model?

By the end of this lesson you can

Explain what query() does, which messages it yields, and how a run ends (success, turn limit, budget limit)
Implement a custom tool with tool() and createSdkMcpServer(), and predict its model-facing name
Trace a tool call through the six-step permission evaluation order and predict whether it runs, is denied or reaches canUseTool
Use PreToolUse and PostToolUse hooks to enforce a write boundary and keep an audit trail
Map every SDK option used in this lesson to its Claude Code equivalent so you can prototype on a subscription

Free · sign in to start

Sign in to keep reading.

Free, no card. We use your account to save progress, quiz scores, bookmarks and occasionally send updates from our team.

12 more sections
3 hands-on exercises
A 7-question quiz
Progress, quiz scores and bookmarks