Skip to content
    Back to the CCA course
    Practice Quiz & Mock Exam

    Claude Certified Architect practice

    Free practice across all five domains, with instant scoring, a per-domain gap analysis, and an explanation for every answer. Pass mark: 70%.

    Preview the question bank (37)
    1. [D1]In the core agentic loop, what does the model produce when it needs information it doesn't have?
    2. [D1]When should you split work across multiple specialised agents instead of one agent with many tools?
    3. [D1]What is the primary purpose of hooks in an agentic workflow?
    4. [D1]For a long-running agent that may crash mid-task, what design best supports recovery?
    5. [D1]In an orchestrator/sub-agent pattern, what is the orchestrator mainly responsible for?
    6. [D2]Why is a tool's description the most important part of its definition?
    7. [D2]A tool fails because a record doesn't exist. What should it return to the model?
    8. [D2]What problem does the Model Context Protocol (MCP) primarily solve?
    9. [D2]When many independent tool calls are needed, which approach improves throughput?
    10. [D3]What is the role of a CLAUDE.md file in a repository?
    11. [D3]When is plan mode preferable to direct execution in Claude Code?
    12. [D3]How can Claude Code be integrated into CI/CD pipelines?
    13. [D3]Why prefer structural file-editing tools over pasting whole files back?
    14. [D4]What most reliably improves the quality of a model's output for a complex task?
    15. [D4]What is the most robust way to get machine-parseable structured output?
    16. [D4]For a large set of independent, non-urgent generations, which API is most cost-effective?
    17. [D4]What does prompt chaining accomplish?
    18. [D5]As a conversation approaches the context window limit, what is a sound strategy?
    19. [D5]When the model is uncertain or the stakes are high, the system should…
    20. [D5]When sources in a RAG context conflict, the model should ideally…
    21. [D1]A long-running agent loop could run indefinitely. What's the right safeguard?
    22. [D1]Why give each sub-agent its own scoped context and tools?
    23. [D1]Where should a human-in-the-loop checkpoint go in an agent that can take irreversible actions?
    24. [D1]Agent actions may be retried after failures. What property should those actions have?
    25. [D1]Best way to constrain what an agent is allowed to do?
    26. [D2]Who is responsible for validating a tool's input arguments?
    27. [D2]You need a capability that a built-in/first-party tool already covers. Best choice?
    28. [D2]An agent has 25 overlapping tools and often picks the wrong one. First fix?
    29. [D3]If a nested directory has its own CLAUDE.md, how does it relate to the root one?
    30. [D3]What are custom slash commands / skills mainly for in Claude Code?
    31. [D3]Running Claude Code in automation (CI). What's the safe practice for changes?
    32. [D4]For deterministic data extraction, which temperature setting is most appropriate?
    33. [D4]How can you increase confidence in a high-stakes answer using the model itself?
    34. [D4]A task has several dependent reasoning steps and keeps failing as one prompt. Better approach?
    35. [D5]A large, stable system prompt is sent on every request at high volume. What cuts latency and cost?
    36. [D5]Your AI service hits provider rate limits under load. Most robust handling?
    37. [D5]How do you keep an AI feature's quality from silently regressing over time?

    Sample Claude Certified Architect Foundations practice questions

    6 free sample questions from the practice bank, with the correct answer and the reasoning.

    1. Question 1 · Domain 1 — Agentic Architecture & Orchestration

      In the core agentic loop, what does the model produce when it needs information it doesn't have?

      • A.A final answer with a disclaimer
      • B.A tool-use request the application executes, returning the result to the model Correct answer
      • C.An error that ends the conversation
      • D.A request for the user to paste the data manually

      Why: The loop is: model reasons → emits a tool-use request → the app runs the tool and feeds the result back → model continues until it can answer. The model never executes tools itself.

    2. Question 2 · Domain 2 — Tool Design & MCP

      Why is a tool's description the most important part of its definition?

      • A.It is shown to end users
      • B.It is the routing signal the model uses to decide when (and whether) to call the tool Correct answer
      • C.It sets the tool's rate limit
      • D.It determines the response format

      Why: The model picks tools based on their descriptions. Vague descriptions cause wrong-tool selection; precise 'when to use / what it returns' wording is what makes routing reliable.

    3. Question 3 · Domain 3 — Claude Code Workflows

      What is the role of a CLAUDE.md file in a repository?

      • A.It stores secrets
      • B.It provides persistent project context/instructions Claude Code reads to follow your conventions Correct answer
      • C.It is the build output
      • D.It replaces the README for users

      Why: CLAUDE.md gives Claude Code durable project-specific guidance (conventions, commands, do/don'ts). Nested CLAUDE.md files layer context by directory.

    4. Question 4 · Domain 4 — Prompting & Structured Output

      What most reliably improves the quality of a model's output for a complex task?

      • A.Vaguer prompts so the model is creative
      • B.Explicit success criteria and, where helpful, a few high-quality examples (few-shot) Correct answer
      • C.Maximising temperature
      • D.Removing all instructions

      Why: Stating explicit criteria and providing a few representative examples steers the model far more reliably than vague prompts or temperature tweaks.

    5. Question 5 · Domain 5 — Context & Reliability

      As a conversation approaches the context window limit, what is a sound strategy?

      • A.Ignore it; the model handles it automatically
      • B.Summarise or prune older turns / externalise state to stay within the window Correct answer
      • C.Lower the temperature
      • D.Switch to a smaller model

      Why: Context is bounded. Summarising/pruning history and externalising durable state keeps the working context relevant and within limits.

    6. Question 6 · Domain 1 — Agentic Architecture & Orchestration

      When should you split work across multiple specialised agents instead of one agent with many tools?

      • A.Never — one agent is always simpler
      • B.When the task spans distinct domains and a single agent's tool list grows large enough to hurt selection reliability Correct answer
      • C.Only when you have more than 100 tools
      • D.Whenever latency matters

      Why: Too many tools on one agent degrades tool-selection accuracy. Splitting specialised work across orchestrated sub-agents keeps each agent's choice space small and reliable.

    What this practice exam covers

    37 questions across 5 exam domains, plus a timed mock exam of 25.

    • Domain 1 — Agentic Architecture & Orchestration
    • Domain 2 — Tool Design & MCP
    • Domain 3 — Claude Code Workflows
    • Domain 4 — Prompting & Structured Output
    • Domain 5 — Context & Reliability

    More practice for this exam

    Claude Certified Architect Foundations: the 386-question bank

    Practice questions written against the published exam domains. Not official exam questions. Not affiliated with or endorsed by Anthropic.

    FAQ

    Is this Claude Certified Architect practice quiz free?

    Yes — taking the quiz is free with no signup, and you can retake it as many times as you like. A free account unlocks the full answer review, per-domain gap analysis and your certificate.

    How many questions are there and what's the pass mark?

    The quick quiz uses the full question bank; the mock exam is 25 domain-weighted questions on a 30-minute timer. The pass mark is 70%.

    Are these the real CCA exam questions?

    No. These are original practice questions written from the publicly documented exam topics. It's an independent study aid and is not affiliated with Anthropic.

    What does the quiz cover?

    All five CCA Foundations domains: agentic architecture, tool design & MCP, Claude Code workflows, prompt engineering & structured output, and context & reliability.

    Does it tell me what to study next?

    Yes — your results include a per-domain gap analysis with links straight to the matching course pages so you can review your weak areas.