Claude Certified Architect – Foundations (CCA-F) exam prep

Free study guide, practice questions and mock exams for Anthropic's Claude Certified Architect – Foundations exam (CCA-F, now also coded CCAR-F). Everything you need, on one page. Bookmark this page. Every lesson, mind map, trap and mock exam is linked from here. The exam tests judgement: each question gives you several sensible options, and you pick the one that fits the situation best.

  1. The exam in one minute
  2. Your study path
  3. Scenarios × domains
  4. The five domains
  5. Mind maps
  6. Trap gym: all 30 traps
  7. Mock exams
  8. Reference pages

1 The exam in one minute

60 questions in 120 minutes. Multiple choice, plus some “pick two”. Each question says how many to select.
4 scenarios drawn from a bank of 6. Each scenario is a system you are helping to design or fix.
Pass mark: 720 on a scale of 100–1,000. Results show your percent correct by domain.
The questions are plain. The hard part is the options: they're all real practices, and only one fits best.

Most wrong answers are good practices used in the wrong place. Three families cover most of them:

Right idea, wrong layer. A prompt where a rule needs code, or code where a prompt is enough.
Fixing the symptom. A retry, a bigger limit or a log, when the cause is somewhere else.
Right content, wrong home. A personal file for a team rule, or an always-loaded file for an occasional task.

2 Your study path

  1. Learn the five domains, biggest first (section 4). Each has lessons, diagrams and a quiz bank.
  2. Read the mind map for each scenario (section 5). Print it, cover the Fix column, and fill it in from memory.
  3. Train on the traps (section 6). Each trap page has a simulator that shows each option playing out, then exam-style practice.
  4. Sit Mock #7, plain style under the 120-minute timer. It reads like the real exam.
  5. Review every miss with “Think it through”. Note the trap name on the option you picked. If the same trap keeps catching you, go back to its trap page.
  6. Keep practising with Mocks #6, #5 and #4, and the per-domain practice sets.

3 Scenarios × domains

Each scenario tests only some of the five domains. Green cells are tested, according to the official exam guide, and each links to the mind map that covers it. Grey cells are not applicable.

ScenarioD1 Agentic
27%
D2 Tools & MCP
18%
D3 Claude Code
20%
D4 Prompts & output
20%
D5 Context
15%
1 Customer Support Resolution Agent ✓Support maps 1–3 ✓MCP maps 3–4 N/AN/A ✓Support maps 4–6
2 Code Generation with Claude Code N/AN/A ✓Claude Code maps 1–3 N/A ✓Claude Code map 5
3 Multi-Agent Research System ✓Multi-agent maps 1–3 ✓Multi-agent map 4MCP map 4 N/AN/A ✓Multi-agent maps 5–6
4 Developer Productivity with Claude ✓Claude Code map 2 ✓Claude Code map 6MCP maps 1–3 ✓Claude Code maps 1, 2, 5 N/AN/A
5 Claude Code for Continuous Integration N/AN/A ✓Claude Code map 4 ✓Extraction maps 1, 4, 5 N/A
6 Structured Data Extraction N/AN/AN/A ✓Extraction maps 1–5 ✓Extraction map 6

You get 4 of the 6 scenarios, so every domain has at least one of its scenarios on your paper. Study all five.

4 The five domains (official order)

DomainWeightWhat it coversLessonsPractice
D1 Agentic Architecture & Orchestration27%The agent loop and stop_reason, subagents, hooks, enforcement vs prompting, handoffsLessons & quizzesD1 practice
D2 Tool Design & MCP Integration18%Tool descriptions, structured errors, tool_choice, MCP servers and resources, built-in toolsLessons & quizzes · MCP skeletonsD2 practice
D3 Claude Code Configuration & Workflows20%CLAUDE.md levels, rules with paths, commands and skills, plan mode, the CLI in CILessons & quizzes · Hands-on labD3 practice
D4 Prompt Engineering & Structured Output20%Explicit criteria, few-shot, JSON schemas via tool use, validation retries, batchesLessons & quizzesD4 practice
D5 Context Management & Reliability15%Case-facts blocks, lost in the middle, escalation, error propagation, review samplingLessons & quizzesD5 practice

5 Mind maps

One printable page per scenario. Each map is a table of “what you see, where the fault is, the fix”, with a yellow runner-up row, the exam-guide line, and one example question.

PageScenariosThe maps
Scenarios × domainsAllThe table in section 3, on its own page
Support agent1When the loop stops · prompt or hook · cleaning tool results · when to escalate · several matches · keeping exact facts
Claude Code2, 4, 5Where an instruction lives · plan or just do it · how to iterate · Claude Code in CI · long or stale sessions · built-in tools
Multi-agent research3Coverage gaps · what a subagent knows · fixed or adaptive · tools per agent · failed subagents · sources and conflicts
MCP1, 3, 4Where the config goes · tool or resource · wrong tool picked · what a failed tool returns
Structured extraction5, 6JSON that always parses · schemas that don't invent · retries · consistency · batch or now · human review

6 Trap gym: all 30 traps

One trap for each task statement in the exam guide. Each page tells a short story, lets you change the real setting in a simulator and watch the effect, and ends with exam-style practice. Open the full trap list →

TrapGuideTrapGuide
D1 Agentic Architecture
Stop reason1.1Hook to clean up data1.5
Coordinator split1.2Fixed chain or Claude decides1.6
Subagent context1.3Resume, fork or start fresh1.7
Prompt or hook1.4, 1.5
D2 Tool Design & MCP
Tool picker2.1tool_choice2.3, 4.3
Tool failure2.2, 5.3MCP: where and what2.4
Too many tools2.3Built-in tools2.5
D3 Claude Code
Where an instruction lives3.1–3.3Examples, tests first or interview3.5
Plan or just do it3.4Claude Code in CI3.6
D4 Prompt Engineering & Structured Output
Explicit criteria4.1Retry with the error4.4
Few-shot examples4.2Batch or now4.5
Schema and nullable fields4.3, 4.4Self-review4.6, 3.6
D5 Context Management & Reliability
Context meter5.1Empty result or failure5.3, 2.2
Who escalates5.2Large codebase5.4
Several customer matches5.2Confidence and sampling5.5
Sources5.6

Some pages cover two traps: “Tool failure” also covers “Empty result or failure”, and “Extraction honesty” covers both schema traps.

7 Mock exams

Four full papers in plain style: 240 questions, each with “Explain simply” and “Think it through”. Take them under the 120-minute timer.

MockUse it for
#7Start here. Reads like the real exam: every option works, and one fits best.
#6 · #5 · #4Three more full papers in the same style, on different situations.
Originals: #7 · #6 · #5 · #4Earlier versions. #7 has a replay simulator for every question. #6 is a stretch: it's harder to read than the real exam.
#1 · #2 · #3Older style. The right answer is often the longest option, so scores here overstate readiness.

8 Reference pages

PageWhen to use it
Exam vs 2026Eight places where the exam's answer differs from today's API and Claude Code. Rule: answer from the guide.
Wire-format referenceWhat API requests and responses look like, with every field marked must know, recognise or skip.
Full study guideAll task statements in one long page.
MCP skeletonsMinimal MCP server and client code, to see the shapes.
Claude Code labTry CLAUDE.md, rules, commands, skills and plan mode yourself.