Skip to content

Module 06 — The Claude API

Level: Beginner → Advanced · Time: ~7 hours (pages) + ~3 hours (labs) · Exam relevance: CCAR-P D1, D2, D3, D4 · CCAO-F D3

This module takes you from your first authenticated request to production-grade API usage: the Messages API and its content blocks, streaming, images and PDFs, the Files API and citations, batch processing, error handling and retries, the cloud platforms that also serve Claude, and how to count tokens and estimate cost.

Every example follows the current API rules in docs/STYLE.md section 1.5 and the official Claude API documentation as of 2026-10-24: model IDs without date suffixes (claude-opus-5-5 by default), no temperature / top_p / top_k, no assistant prefill, adaptive thinking tuned with effort, and a stop_reason check before using any output. Python comes first; cURL appears where it helps you see the raw HTTP shape.

Learning objectives

By the end of this module you can:

  1. Set up a Console account, create and protect an API key, and install the Python SDK.
  2. Build Messages API requests with roles, system prompts and content blocks, and handle every stop reason, including refusals.
  3. Stream responses and choose between the streaming helpers.
  4. Send images and PDFs, reuse files with the Files API, and get verifiable citations.
  5. Process large workloads asynchronously with the Message Batches API.
  6. Handle errors with a typed exception chain, respect rate limits and design retries.
  7. Explain the cloud platform options (Claude Platform on AWS, Amazon Bedrock, Google Vertex AI, Microsoft Foundry) and their client classes.
  8. Count tokens, read the usage object and estimate cost.

Page order

# Page What you learn
1 Getting started with the API Console, workspaces, API keys in environment variables, SDK install, first request
2 The Messages API Request anatomy, roles, content blocks, stop reasons, refusals, effort, multi-turn
3 Streaming Server-sent events, SDK helpers, long outputs, handling stream errors
4 Vision, documents and citations Images, PDFs, the Files API, the citations feature
5 Batch processing The Message Batches API, limits, result handling, cost
6 Errors, rate limits and retries HTTP errors, typed exceptions, rate-limit headers, backoff
7 Cloud platforms Claude Platform on AWS, Bedrock, Vertex AI, Foundry; client classes and feature differences
8 Token counting and cost count_tokens, usage fields, pricing levers, cost estimation

Prerequisites

Exam relevance

Objective IDs come from 16-certification-tracks/registry.yaml. The Professional exam is architecture-focused: it tests whether you can choose and justify API features, not whether you remember method names. Foundations candidates mainly need page 1 (what the API is) and the cost concepts on page 8.

Certification Objectives covered Where
CCAR-P CCAR-P-1.2 Design end-to-end architectures Pages 2, 5, 7
CCAR-P CCAR-P-2.1 Select models based on trade-offs Pages 2, 8
CCAR-P CCAR-P-2.4 Optimise context windows and token usage Pages 4, 8
CCAR-P CCAR-P-3.2 Analyse authentication and authorisation Pages 1, 7
CCAR-P CCAR-P-3.3 Accuracy–latency trade-offs Pages 2, 3, 5
CCAR-P CCAR-P-3.4 Observability at scale Pages 6, 8
CCAR-P CCAR-P-3.7 Choose the integration mechanism Pages 1, 7
CCAR-P CCAR-P-4.5 Optimise token usage, latency and cost Pages 5, 8
CCAR-P CCAR-P-5.4 Regulatory compliance (data residency, platform choice) Page 7
CCAR-P CCAR-P-7.3 Debugging and operational issue resolution Page 6
CCAO-F CCAO-F-3.3 Match model choice to cost, speed and quality Page 8

Hands-on labs

Cost: requires API credits (live mode). All labs run offline in mock mode by default.

Quick reference

Next module

Continue to Module 07 — Tool Use and Function Calling, where Claude starts calling your code.