Anthropic's own guidance is blunt: without hands-on experience with Claude Code, the Agent SDK, the Claude API and MCP, you will probably not pass. So the plan is to build. Budget roughly 30–37 hours spread over four weeks — spread it out, do not cram.
Prerequisites
- Python 3.11+
- An Anthropic API key
- Claude Code installed
- Willingness to write code
Four free courses first (Anthropic Academy)
- Claude with the Anthropic API — about 8–10 hours; the most important single course.
- Claude Code in Action — hands-on agentic coding.
- Introduction to Agent Skills.
- Introduction to Model Context Protocol — Tools versus Resources, servers.
Five things to build before you sit
- A validation-retry loop. Extract structured data with
tool_useand a JSON schema, validate it, and on failure retry with the document, the failed extraction and the specific error. - A multi-agent system with explicit context passing. A coordinator spawns two subagents with deliberately different context; compare the outputs.
- A Claude Code project with a proper CLAUDE.md hierarchy — project standards, personal preferences, a path-scoped rule for test files and one skill with
context: fork. - Claude Code in a CI/CD simulation. Run it with
-p, add--output-format json, and parse the result. - An MCP server with a Tool and a Resource. Connect it to Claude Code and notice how each is used.
The four weeks
| Week | Focus |
|---|---|
| 1 — Foundations | All four courses; get the vocabulary cold. |
| 2 — Build all five | The most important week. Hands on keyboard. |
| 3 — Scenarios and sample questions | Work through all six scenarios; use Anthropic's sample questions to find gaps. |
| 4 — Drill and capstone | Drill the common distractors, build a capstone, do a final readiness check. |
What is not on the exam
Do not spend time on: fine-tuning or training models, API authentication or billing, cloud-provider configuration, Constitutional AI or safety training, vector databases and embeddings, computer use or browser automation, vision, the streaming API, rate limits and pricing math, OAuth and key rotation, or prompt-caching internals (just know it exists).
Prefer guided preparation?
Our two-day workshop builds six labs with a trainer, and every participant gets a practice portal. Otherwise, the free community is a good place to compare notes.