Is the CCA Exam Multiple Choice? Format, Types, and a Sample

Yes. The CCA Foundations (CCA-F) exam is entirely multiple choice: 60 selected-response questions in 120 minutes, mixing single-answer multiple-choice items with multiple-response items where you pick a stated number of options. There is no coding, no lab, and no free-text answer. Confirm the current format on the official Anthropic Academy or Pearson VUE page before you register.

“Multiple choice” undersells it, though. Almost every item is a scenario, and the four options are usually all things a reasonable engineer might do. This post explains the two question types, the recurring “least bad design” framing, and walks through a sample question written in that style.

What question types appear on the CCA-F exam?

Two, and the difference matters mechanically:

Type	Instruction	How it is scored
Multiple choice (single answer)	“Which one of the following…”	One correct option out of four
Multiple response	”Select two” / “Choose all that apply” style, with the count stated	You must select the specified number; Anthropic has not published whether partial credit applies, so assume all-or-nothing

The instruction line tells you which type you are on. The most avoidable mistake on the exam is treating a select-two item as a single-answer one, or selecting three when two were asked for. In Pearson VUE’s interface, multiple-response items usually use checkboxes rather than radio buttons, which is a visual cue, but read the stem anyway.

Why does everyone call the questions “scenario-based”?

Because a typical stem is not “What does tool_choice: any do?” but rather:

A team runs a support agent that calls three tools. They add twelve more integrations and routing accuracy drops sharply. They cannot remove any integration. Which change is most likely to restore accuracy?

The knowledge being tested is the same (tool count degrades selection, descriptions and scoping fix it), but you have to recognise the mechanism inside a story and then pick the option that respects the constraint “cannot remove any integration.” That is the exam’s core move: it gives you a goal, adds a constraint, and asks for the least bad design under that constraint. Options that would be great without the constraint are the distractors.

Practical consequence: read the constraint sentence first, then evaluate options against it. Two options usually violate it outright and can be eliminated in seconds.

What does a CCA-F question actually look like?

Below is a sample question written by us in the style of the exam. It is not an official item and does not appear on the real test.

Sample. A research assistant built on the Agent SDK dispatches five sub-agents to gather sources, then a coordinator writes a report. Each sub-agent returns its full reasoning transcript, roughly 3,000 tokens, and the coordinator’s context fills up before it can synthesise all five. The team needs every source found by every sub-agent to appear in the report. Which change is the least bad given that requirement?

• A. Reduce the number of sub-agents from five to three so the transcripts fit.
• B. Have each sub-agent return a structured summary of its findings (source, claim, confidence) instead of its full transcript.
• C. Increase the coordinator’s system prompt to instruct it to “be concise” when synthesising.
• D. Ask the coordinator to summarise each transcript progressively as it reads them.

Reasoning. A drops coverage, which the requirement forbids. C does nothing about input size; the problem is inbound context, not outbound verbosity. D preserves the token pressure and risks losing specifics through repeated summarisation. B shrinks each sub-agent’s return by an order of magnitude while keeping every source, so it satisfies the constraint with the fewest side effects. Answer: B. The underlying idea is covered in Sub-agent structured output and Context passing best practices.

Notice that A, C, and D are not absurd. Each is something a team might genuinely try. That is what makes the exam feel harder than a recall test.

How is the exam scored and reported?

Answers are converted to a scaled score from 100 to 1000, and you need 720 to pass. Score reports typically show pass/fail, the scaled score, and per-domain performance as informational bands. There is no published penalty for wrong answers, so never leave an item blank. More detail in CCA exam passing score explained.

How should the format change the way I prepare?

• Practise both item types. Our practice hub includes single-answer and multi-select questions, each with an explanation, so the choose-two instruction stops surprising you.
• Drill elimination, not recall. For each practice item, say why each wrong option is wrong. The exam rewards recognising the violated constraint faster than it rewards knowing more facts.
• Read explanations for questions you got right. Getting the right answer for the wrong reason is how people pass mocks and fail the exam.
• Take timed mocks. The 60-question mock is sampled by official domain weight and uses a 90-minute clock, tighter than the real 120, so the format and pace stop being novel.

Next step

Take the free mock exam to see the two question types under a clock, then work through the Domain 1 study guide. Domain 1 carries the most scenario-style items on the exam, and it is where the “least bad design” reflex pays off first.