Anthropic launched its first official technical certification on March 12, 2026 — the Claude Certified Architect (CCA) Foundations exam. It's a proctored, scenario-based exam designed for engineers building production applications with Claude.
There's almost no prep material out there yet. So I put together a complete study guide: a 7-week roadmap and a curated reading list of 38 official docs, mapped directly from Anthropic's official exam guide PDF.
What the exam actually tests
The CCA Foundations has 60 questions across 5 domains:
| Domain | Weight |
|---|---|
| Agentic Architecture & Orchestration | 27% |
| Claude Code Configuration & Workflows | 20% |
| Prompt Engineering & Structured Output | 20% |
| Tool Design & MCP Integration | 18% |
| Context Management & Reliability | 15% |
The questions are scenario-based — you get a realistic production context and have to choose the best architectural decision. It's not about memorizing API syntax. It's about knowing why you'd use programmatic hooks over prompt instructions, or when the Batch API is appropriate vs a blocking API call.
The 7-week roadmap
Phase 1 — Claude API & core concepts (Week 1–2)
Start here regardless of your background. The agentic loop pattern — send request → check stop_reason → execute tool → return result → repeat — is the mental model behind 65% of the exam. Everything else builds on it.
Phase 2 — Claude Code hands-on (Week 2–3)
Install Claude Code and run it on a real project. The exam tests CLAUDE.md hierarchy, .claude/rules/ glob patterns, custom slash commands, plan mode vs direct execution, and the -p flag for CI/CD pipelines. Reading docs isn't enough here — you need to have used it.
Phase 3 — MCP & tool design (Week 3–4)
Learn the Model Context Protocol: how tools are described, how errors are structured (isError, isRetryable, errorCategory), and how to configure MCP servers in .mcp.json. The biggest source of exam mistakes here is not understanding how tool descriptions drive Claude's tool selection.
Phase 4 — Prompt engineering & structured output (Week 4–5)
Few-shot prompting, validation-retry loops, the Message Batches API (50% cost savings, 24h window — only for latency-tolerant jobs), and multi-pass review architecture. The exam heavily tests judgment calls on which technique to use when.
Phase 5 — Multi-agent systems & context management (Week 5–6)
The biggest domain at 42% combined. Coordinator-subagent patterns, the Task tool, PostToolUse hooks, context window optimization, escalation patterns, error propagation across agents, information provenance. Build the multi-agent exercise from the official exam guide — it covers most of this in one go.
Phase 6 — Final prep (Week 6–7)
Work all 12 sample questions from the exam guide, build the 4 prep exercises, review the out-of-scope list (fine-tuning, auth, vision, streaming — don't waste time here), and take the official practice exam before sitting the real one.
Key concepts that trip people up
These show up as wrong answers specifically because they seem right:
- Using prompt instructions to enforce critical business rules — wrong, use programmatic hooks
- Parsing natural language to determine loop termination — wrong, check
stop_reason - Self-review in the same session — wrong, use an independent review instance
- Giving an agent 18 tools — wrong, 4–5 tools max for reliable selection
- Sentiment-based escalation — wrong, sentiment ≠ complexity
The full interactive guide
I built the complete prep resource as an interactive guide — 7-week visual roadmap + all 38 docs organized by domain with must-read / good-to-read / reference labels.
👉 How to Pass the Claude Certified Architect Exam
It's free, no signup required, and everything links directly to the official Anthropic and MCP documentation.
The cert is brand new and the window to be early is open right now. Happy to answer questions in the comments if you're studying for it.
Top comments (0)