Skip to main content

Claude API tutorial

The fastest useful way to learn the Claude API is to build one small service that has to make real engineering tradeoffs. This tutorial uses a customer-support triage API: it classifies an incoming ticket, decides when a person must review it, and measures whether the result is reliable.

You need Node 20+, an Anthropic API key, and about 20 minutes for the first working request. The complete course takes about four hours and costs approximately $2–4 in API usage.

What you will build​

  • A typed ticket classifier using structured outputs
  • A tool-using resolution step that checks orders and policy rather than trusting customer claims
  • A streaming reply endpoint with cost accounting
  • A cached policy prefix and an eval suite to prove changes help

Make your first Claude API request​

Install the SDK and create a request with a model and a message:

import Anthropic from "@anthropic-ai/sdk";

const anthropic = new Anthropic({ apiKey: process.env.ANTHROPIC_API_KEY });

const response = await anthropic.messages.create({
  model: "claude-sonnet-4-5",
  max_tokens: 200,
  messages: [{ role: "user", content: "Classify this support ticket: My jacket zipper broke." }],
});

That request proves connectivity, but it is not a service contract. A production caller needs predictable fields, evidence for consequential decisions, cost visibility, and a way to detect regressions. Those are the four layers the course adds.

Common mistakes​

  • Treating a valid HTTP response as proof the answer is correct. Measure representative cases before relying on it.
  • Asking for JSON in a prompt and parsing it by hand. Use structured outputs when code depends on the response shape.
  • Building an agent loop without a maximum iteration count or total-usage accounting.
  • Optimizing a prompt before recording a baseline.

Continue with the hands-on course​

Start with Lab 1: your first call, then take the labs in order. If you are new to the vocabulary, begin with Start from zero. For the full design context, read the Northwind scenario.