Study guide and training for exam CCAR-F

Claude Code Architect

Built from Anthropic's own exam guide for the Claude Certified Architect, Foundations certification. Every one of its 30 task statements is here, rewritten as study notes, plus practice questions set in its six scenarios, decision drills, a stop_reason loop simulator, flashcards and hands-on labs. Rate yourself as you go and the panel on the right tells you what to study next.

60items, single and multiple response
120minutes
720to pass, scaled 100 to 1000
4 of 6scenarios per sitting

Source: Claude Certified Architect, Foundations Exam Guide, version 1.0, effective July 2026, from the Anthropic Partner Academy. The notes and every practice question here are written for this page; the guide's own twelve sample questions are in the PDF. Claude Code details checked against code.claude.com on 27 September 2026.

The exam in one screen

CredentialClaude Certified Architect, Foundations. Exam code CCAR-F.
TestsTradeoff decisions across Claude Code, the Claude Agent SDK, the Claude API and MCP, framed by production scenarios.
Format60 items, multiple choice and multiple response (each item says how many to pick). 4 scenarios drawn at random from 6. 120 minutes, proctored online or at a Pearson VUE centre.
ScoringCriterion referenced, scaled 100 to 1000, pass at 720. The score report adds percent correct per domain, which does not decide pass or fail.
Cost and access$125 per attempt, registered through the Anthropic Partner Academy (partner tier discounts apply at checkout).
RetakesWait 14 days after a first fail, 30 after a second, 90 after a third. At most four attempts in a rolling twelve months.
Validity12 months. On time renewal is a free, unproctored assessment on the Partner Academy; a lapsed credential means the full exam again.
CandidateA solution architect with roughly six months or more of hands-on work with the Claude API, Agent SDK, Claude Code and MCP.

Domains

Click one to open its task statements

The six scenarios

Four appear in any sitting

Out of scope, so do not spend time on it

  • Fine tuning or training models; Claude's internals, RLHF, Constitutional AI
  • API authentication, billing, rate limits, pricing maths, key rotation, OAuth
  • Hosting MCP servers: infrastructure, networking, containers
  • Cloud provider specifics (AWS, GCP, Azure)
  • Computer use, vision, streaming and server sent events
  • Embeddings and vector databases, benchmarks, tokenisation
  • Prompt caching beyond knowing it exists

How the questions think

  • Guaranteed, must, financial or compliance in the stem points to code: a hook or a prerequisite gate, not better wording.
  • Most effective first step means the cheapest fix at the root cause. Better tool descriptions come before routers and classifiers.
  • Answers that add infrastructure (a classifier, a routing layer, a bigger model) are usually the over engineered distractor.
  • Blame the component the logs point at. If every subagent succeeded, the coordinator's plan is the suspect.
  • Self reported confidence and sentiment are unreliable signals for escalation.
  • Suppressing an error and killing the whole workflow are both wrong; return structured context.