Questions about the CCAR-F exam

It is Anthropic's first official technical credential, aimed at solution architects who design and build production applications with Claude. The exam tests judgement rather than recall across Claude Code, the Claude Agent SDK, the Claude API and the Model Context Protocol (MCP), using scenario-based questions drawn from real customer use cases.

The official exam code in the published exam guide is CCAR-F. You will also see CCA-F and CCAF used widely — on course listings, in forums and in this site's own domain name — and all three refer to the same certification. Anthropic's other codes follow the same pattern: CCAO-F for Associate, CCDV-F for Developer, CCAR-P for Architect Professional.

60 items in 120 minutes. Items are multiple-choice and multiple-response, and each one states how many responses to select — so "choose two" is information rather than a hint. Questions are grouped under 4 scenarios drawn at random from a published bank of 6.

A scaled score of 720 on a 100–1000 scale. Scaling adjusts for the difficulty of the particular items you were given, so it is not a straight percentage of questions answered correctly. Your score report also breaks down percent-correct by domain.

$125 USD. Registration goes through the Anthropic Partner Academy, and the exam itself is delivered by Pearson VUE — either online proctored or at a test centre. You can cancel or reschedule up to 24 hours before your appointment; changes inside 24 hours forfeit the fee.

12 months from the date it is awarded. Renewing on time means reviewing what has changed and completing a free, non-proctored assessment on the Anthropic Partner Academy — there is no fee for an on-time renewal. If the credential lapses you must retake the full exam at full price.

You can retake it, with waiting periods that increase per attempt: 14 days after the first failure, 30 days after the second, 90 days after the third. You may sit the exam up to four times in a rolling twelve-month period.

Agentic Architecture & Orchestration (27%), Tool Design & MCP Integration (18%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%) and Context Management & Reliability (15%). Study time is best spent in that order — the heaviest domain is worth nearly twice the lightest.

The exam guide publishes a list, and it is worth reading before you over-prepare. Fine-tuning, model internals, Constitutional AI and RLHF, embeddings and vector databases, computer use, vision, streaming implementation, rate limits and pricing, OAuth and key rotation, cloud provider configuration, and MCP server hosting and infrastructure are all named as out of scope.

The exam guide describes an ideal candidate with roughly six months of hands-on experience building with the Claude API, Agent SDK, Claude Code and MCP. If you have that background, plan on 15–20 hours with this curriculum. If you are newer to Claude, budget 30–40 hours and spend the extra time on the build exercises rather than on re-reading.

Agentic Architecture & Orchestration, at 27% of the exam. It is the heaviest domain by some margin, and the concepts in it — agentic loops, orchestration patterns, context passing — recur inside the other four. Studying in blueprint-weight order puts your time where the marks are.

There are 318 items in total, including the 12 official sample questions verbatim. The rest are written against the official task statements, with an explanation for every option rather than just the correct one — because the exam is a judgement test and knowing why the other three answers fail is most of the skill.

It samples 60 questions in the official blueprint proportions and reports a scaled score against the real pass mark of 720. Anthropic has not published its scaling function, so no third party can reproduce it. Treat the number as a relative gauge and act on the per-domain breakdown underneath it.

The exam asks which design you would choose under a constraint, which is a different skill from recognising vocabulary. Reading gets you to the point of knowing that hooks exist; having once written the wrong hook and watched it fire at the wrong time is what lets you answer quickly under a clock. The build exercises in each lesson exist for that reason.

Not new material. The cheat sheets are built for exactly this — one dense page per domain, tables and terse bullets, printable. Skim the divergence log too, so a question written against the July 2026 guide does not catch you answering from this week's documentation.

Yes. Every lesson, every practice question, the full mock exam, the flashcards, the cheat sheets and the glossary are open with no login, no paywall and no locked content. There is an optional tip link for anyone who wants to support the upkeep, and that is the entire business model.

No. It is an independent study resource, not affiliated with, endorsed by or sponsored by Anthropic, PBC. Claude and Claude Certified Architect are trademarks of Anthropic, PBC. Register for the actual exam through the Anthropic Partner Academy.

The official exam guide, Anthropic's product documentation, and Anthropic's engineering write-ups. Every claim in a lesson's Deep Dive carries a link to the source it came from, so you can check any statement rather than take it on trust. A build-time check fails if a Deep Dive section is missing its source line.

Both are recorded, and the exam guide answer is flagged as the one to give in the exam. Software ships faster than certification blueprints, so this happens more than you would expect; the divergence log lists every case found so far.

No accounts, no login, no email collection, and no cookies. Your study progress is kept in your own browser's local storage and is never sent anywhere — there is no server to send it to. Analytics are cookieless and aggregate-only.

No, and you should be sceptical of anyone claiming otherwise — the real item bank is confidential and redistributing it breaches the exam agreement. The twelve official sample questions published in the exam guide appear here verbatim, and everything else is written against the same task statements.

It is the architect track's entry point. The programme also includes CCAO-F (Associate – Foundations), CCDV-F (Developer – Foundations) and CCAR-P (Architect – Professional). Foundations-level credentials have no formal prerequisites, so you can sit CCAR-F without holding any of the others.

No formal prerequisite. The exam guide describes an ideal candidate with 6+ months of practical experience building with the Claude API, Agent SDK, Claude Code and MCP, but that is a description of readiness rather than a gate — nothing stops you booking it earlier.

That depends on what you want from it. It is a new credential, so it has little hiring signal yet; what it does have is a genuinely well-constructed blueprint. Preparing for it is a structured way to cover production Claude architecture properly, and the exam itself is a fair test of whether you have. If you want the study, the credential is a useful forcing function. If you only want the badge, wait and see how the market treats it.

No. It is closed-book and proctored, whether you sit it online or at a test centre. You must stay in view of the webcam for the whole session if testing online, and permitted items — scratch paper, for instance — are specified by Pearson VUE.