The Certification Path

Not all Skilljar courses carry equal weight for the exam. Here is a prioritized breakdown of the courses most relevant to each exam domain, so you can allocate your time effectively.

The exam tests a specific set of technologies. For each one, you need both conceptual understanding (what it is, when to use it) and practical fluency (how to configure and use it). Here is what is in scope.

• ▸Claude API (Messages API) — The core HTTP API for interacting with Claude. You must understand the message format, system prompts, multi-turn conversations, tool_use blocks, stop reasons, and how to handle API responses. This is the foundation everything else builds on.
• ▸Claude Agent SDK — The official SDK for building agentic applications. Know how to define agent loops, register tools, implement handoffs between agents, configure guardrails, and handle lifecycle events (hooks). Understand the difference between single-agent and multi-agent architectures.
• ▸MCP (Model Context Protocol) — The protocol for connecting Claude to external tools and data sources. Understand the server/client architecture, tool schema definitions, resource types, and transport layers (stdio, HTTP). Know when to use MCP versus direct tool_use.
• ▸Claude Code — Anthropic's CLI tool for code-aware AI assistance. Master CLAUDE.md file structure, the settings hierarchy (project vs user vs enterprise), slash commands, hooks (pre/post), non-interactive mode for CI/CD, and repository conventions.
• ▸Message Batches API — The API for processing multiple requests efficiently. Know when batch processing is appropriate (high volume, latency-tolerant workloads) versus when to use individual requests.
• ▸JSON Schema — The standard for defining structured output formats. Understand required fields, type definitions, nested objects, arrays, enums, and how Claude uses schemas to produce validated output.
• ▸Pydantic — Python's data validation library used with Claude for structured output. Know how to define models, use field types and validators, handle validation errors, and integrate Pydantic with Claude API responses.

Knowing what is out of scope is just as valuable as knowing what is in scope. The following topics will not appear on the exam. If you encounter an answer choice that references one of these, it is almost certainly a distractor.

• ▸Fine-tuning — Claude does not support fine-tuning. Any answer mentioning 'fine-tune the model' is wrong.
• ▸Billing & pricing — You will not be asked about token costs, pricing tiers, or usage limits.
• ▸Cloud provider configs — AWS Bedrock setup, GCP Vertex AI configuration, and Azure deployment are not tested.
• ▸Embeddings — While Anthropic offers embedding capabilities, they are outside this exam's scope.
• ▸Computer use — The computer use beta (controlling desktop applications) is not covered.
• ▸Vision — Image input and multimodal analysis are not tested on this exam.
• ▸Streaming — Server-sent events, streaming token handling, and real-time response rendering are out of scope.
• ▸Rate limiting — Throttling strategies, retry-after headers, and quota management are not tested.
• ▸OAuth — Authentication protocols and identity management are outside the exam boundary.

This course is structured to map directly to the five exam domains. Each topic covers one domain in depth, with articles that explain concepts, provide working examples, and include scenario-based exercises. Here is how to work through it effectively.

How long you need depends on your starting experience. Here are three profiles with recommended timelines.

The Claude Cookbooks repository at github.com/anthropics/claude-cookbooks is the most underutilized exam preparation resource. It contains working Jupyter notebooks and Python scripts that demonstrate the exact patterns tested on the exam. Here is how to get maximum value from it.

• ▸Clone the repo and run notebooks locally — reading code on GitHub is not the same as executing and modifying it
• ▸Focus on these cookbook categories: tool use (tool schemas, handling tool results), agents (agent loops, multi-step workflows), and structured output (JSON extraction, Pydantic validation)
• ▸After running each notebook, modify it: change the tool schemas, alter the system prompt, swap the orchestration pattern, and observe how behavior changes
• ▸Break things intentionally: remove error handling, exceed context limits, send malformed tool results — understanding failure modes is essential for the exam
• ▸Compare cookbook implementations against what you learn in this course — the cookbooks show one way to solve a problem, this course explains why that approach was chosen over alternatives