Free CCA Practice Questions: 1,374 CCA-F Questions by Domain

This site offers 1,374 free CCA-F practice questions covering all five exam domains, in two modes: a per-domain mode that serves one random question per knowledge point with no timer, and a 60-question, 90-minute mock exam sampled by official domain weight with a 72% target. Both track your mistakes so review is targeted, and both are free.

Below: what separates a useful CCA practice question from a trivia question, how the question bank is organised, how to use the two modes, and three original sample questions. Exam facts (60 questions, 120 minutes, scaled 720 pass) are as of mid-2026; confirm on the official Anthropic / Pearson VUE page before you register.

What does a good CCA practice question look like?

The real exam is scenario-based: a short description of a system, a symptom or constraint, and four designs, of which you pick the least bad. A practice question that helps you should therefore have:

• A scenario with a constraint that matters. “The action is irreversible”, “latency must not increase”, “the team shares the config through git”. If you can answer without reading the constraint, the question is testing recall, not judgment.
• Distractors that are real anti-patterns. Wrong options should be things teams actually do: add a sentence to the prompt, retry more, add more tools. See the anti-patterns the exam punishes.
• An explanation that names the mechanism. Why the right option guarantees the behaviour (hook, schema, structured error, scoped tools) and why each wrong option does not.
• A link back to a knowledge point, so a miss becomes ten minutes of reading rather than a shrug.

Questions that ask for a definition (“what does MCP stand for”) or a number (“what is the passing score”) are fine for warm-up but do not resemble the exam.

How are the 1,374 free questions organised?

Every question is tagged to one of the 165 study-guide knowledge points, which sit inside the five domains:

Domain	Weight	Practice set	Study guide
D1 Agentic Architecture & Orchestration	27%	D1 practice	Domain 1
D2 Tool Design & MCP Integration	18%	D2 practice	Domain 2
D3 Claude Code Configuration & Workflows	20%	D3 practice	Domain 3
D4 Prompt Engineering & Structured Output	20%	D4 practice	Domain 4
D5 Context Management & Reliability	15%	D5 practice	Domain 5

The bank includes multiple-choice, multiple-response and a small number of true/false items so you see the answer formats you will meet. The practice hub is the entry point.

Per-knowledge-point mode or timed mock: which should I use?

Per-knowledge-point (domain) mode picks one random question for each knowledge point in the domain, with no time limit and immediate explanations. Use it while you are studying a domain: read the article group, then run the domain set, then re-read what you missed. Because it re-samples, running it again gives you different questions on the same knowledge points, which is how you find out whether you learned the idea or memorised one item.

Timed mock exam serves 60 questions in 90 minutes, sampled by the official weights (about 16 D1, 11 D2, 12 D3, 12 D4, 9 D5), with a 72% target. Use it as a checkpoint at the end of each study week and as a rehearsal in the final days. The 90-minute limit is tighter than the real 120 on purpose.

A sensible split, from the 30-day plan: domain mode on weekdays, one mock each weekend.

How should I review mistakes?

The mistakes list is where the score comes from. For each wrong answer:

1. Read the explanation and the linked knowledge-point article.
2. Write one sentence: “I chose X; it fails because under constraint Y it does not guarantee Z.” If you cannot write that sentence, you have not finished the review.
3. Tag the miss as either knowledge (did not know the mechanism) or reading (missed the constraint). Reading misses are fixed by slowing down on the stem; knowledge misses by re-reading the article group.
4. Re-run the domain set a few days later and confirm the knowledge point is now green.

Skip re-doing the exact same question you just saw; that tests memory of the item, not the idea.

Three sample CCA practice questions

These are original samples written for this post, in the exam’s style. Answers follow each one.

Sample 1 (D1). A coordinator agent spawns a sub-agent to check a customer’s refund eligibility. The sub-agent keeps asking which customer it should look up. What is the most likely cause?

• A. The sub-agent’s model is too small.
• B. The sub-agent starts with an empty context and the coordinator did not pass the customer identifier.
• C. The allowedTools list is too long.
• D. The coordinator should have used a hook.

Sample answer: B. Sub-agents do not inherit the parent’s conversation; anything they need must be passed explicitly. See Sub-Agents See Nothing.

Sample 2 (D2). An agent with a get_order tool retries five times whenever the tool fails, including when the failure is a permissions error. Which change fixes this with the least added complexity?

• A. Increase the retry limit so transient errors eventually succeed.
• B. Add “do not retry permission errors” to the system prompt.
• C. Return isError: true with a classified error type (for example permission_denied vs timeout) so the agent can decide whether to retry.
• D. Split get_order into five separate tools.

Sample answer: C. Classified errors let the model distinguish retryable from terminal failures; prompt text alone does not, and a longer retry limit makes it worse. See Error Type Classification.

Sample 3 (D5). A support agent’s summarised history turns “refund $127.50 to card ending 4421” into “customer requested a refund”, and a later turn refunds the wrong amount. What is the least bad fix?

• A. Use a model with a larger context window and stop summarising.
• B. Keep a persistent, structured case-facts block that survives summarisation and is re-injected each turn.
• C. Ask the customer to repeat the amount every turn.
• D. Lower the temperature.

Sample answer: B. Summaries lose specifics; a structured facts block preserves them at low token cost. See Progressive Summarization and Case Facts.

Notice the shape: one option adds prompt text, one adds raw capacity, one is a real mechanism. That is what the exam is really testing.

Next step

Start with the domain set for whatever you are studying this week, then take the free 60-question mock exam to see per-domain scores. If you have not chosen a starting point, Domain 1 is the largest slice of the exam and the best place to begin.