2026 CCAR-F (CCA-F) Practice Test: Free Claude Certified Architect – Foundations Questions

Reviewed by the FlashGenius certification content team · Last updated: August 13, 2026 · Aligned with Anthropic's Claude Certified Architect – Foundations exam objectives

Prepare for Anthropic's Claude Certified Architect – Foundations exam — officially coded CCAR-F, widely abbreviated CCA-F or CCAF — with a 250+ question practice bank. The sample questions below are free with no registration, every account gets 10 free practice questions per day, and instant scoring with detailed explanations comes standard. Covers all 5 official exam domains, including agentic architecture, Claude Code workflows, prompt engineering, tool design and MCP integration, and context management.

Free vs paid CCA-F prep: several practice platforms charge $25 or more for Claude Certified Architect question banks. FlashGenius gives you 10 free questions daily from the full 250+ CCAR-F bank — and Premium at $14.99/month unlocks unlimited access, full exam simulation, and smart review of your weak domains. That's a fraction of the $125 exam fee.

What to Expect on the CCAR-F Exam

CCAR-F is Anthropic's certification for practitioners who design and build production systems with Claude. It validates your ability to architect agentic systems, choose orchestration patterns, design tools and MCP integrations, configure Claude Code workflows, engineer prompts with structured output, and manage context and reliability in production. Candidates should be comfortable reading and reasoning about code that uses the Claude API, the Claude Agent SDK, and MCP servers — the exam emphasizes architectural judgment rather than writing code from scratch.

60 Questions
120 min Exam Time
720/1000 Passing Score
$125 USD Exam Fee

The exam uses a scaled scoring model on a 100–1,000 scale, so you need a scaled score of 720 or higher to pass — the exact number of questions you must answer correctly varies with question difficulty. All questions are scenario-based multiple choice, giving you an average of two minutes per question.

CCAR-F Exam Domains

The CCAR-F exam is organized into five weighted domains. Agentic Architecture & Orchestration carries the most weight at 27%, followed by Claude Code Configuration & Workflows and Prompt Engineering & Structured Output at 20% each.

Domain 1: Agentic Architecture & Orchestration (27%)

Key topics: agent loops, workflows vs agents, orchestrator–worker patterns, subagent delegation, and multi-agent system design.

Practice Agentic Architecture & Orchestration Questions

Domain 2: Claude Code Configuration & Workflows (20%)

Key topics: CLAUDE.md configuration, slash commands, hooks, permissions, headless/CI usage, and team-wide settings.

Practice Claude Code Configuration & Workflows Questions

Domain 3: Prompt Engineering & Structured Output (20%)

Key topics: system prompts, XML tags, few-shot examples, chain-of-thought, prefilling, and enforcing JSON and structured output.

Practice Prompt Engineering & Structured Output Questions

Domain 4: Tool Design & MCP Integration (18%)

Key topics: tool schemas and descriptions, MCP servers and clients, resources vs tools, transports, and tool-use error handling.

Practice Tool Design & MCP Integration Questions

Domain 5: Context Management & Reliability (15%)

Key topics: context window management, compaction, prompt caching, retrieval, guardrails, evals, and human-in-the-loop review.

Practice Context Management & Reliability Questions

10 Free CCAR-F Sample Questions with Answers

Each question below includes 4 answer options, the correct answer, and a detailed explanation. These are real questions from the FlashGenius CCAR-F question bank — two from each of the 5 exam domains.

Sample Question 1 — Agentic Architecture & Orchestration

A support agent calls several account tools before composing a customer response. The current application stops whenever Claude's text contains the word "completed," causing it to terminate before some requested tools run. Which implementation BEST improves reliability?

  1. A. Continue until two consecutive responses contain no tool names.
  2. B. Execute requests on tool_use, append matching tool results, and continue until end_turn. (Correct answer)
  3. C. Begin a new conversation after each tool call to reduce context size.
  4. D. Require Claude to end every final response with a unique completion phrase.

Correct answer: B

Explanation: Correct answer (B): The application should treat stop_reason as the protocol-level control signal. On tool_use, it executes the requested tool, appends the correctly associated tool result to conversation history, and sends the updated conversation back to Claude. It terminates the current loop on end_turn. This matters in production because free-form wording is variable, whereas protocol-level signals and preserved tool history support reliable multi-step execution. Why the other options are wrong: - Option A: Tool names in generated text are not authoritative execution signals, and waiting for two responses could add unnecessary latency. - Option C: Fresh conversations discard the tool requests and prior observations needed to continue the task coherently. - Option D: A completion phrase still relies on nondeterministic natural-language compliance rather than the API's stop reason.

Sample Question 2 — Agentic Architecture & Orchestration

A research agent can search regulatory filings, query market data, and retrieve internal reports. The correct next source depends on evidence returned by the previous tool, but production policy requires a maximum of eight tool calls. Which orchestration approach is MOST appropriate?

  1. A. Always call the three tools in a fixed order and stop after one pass.
  2. B. Let Claude invoke tools without retaining results or enforcing a call limit.
  3. C. Let Claude select each next tool while the host preserves results and enforces the limit. (Correct answer)
  4. D. Ask Claude to list every tool it might need before the host executes any.

Correct answer: C

Explanation: Correct answer (C): Model-driven tool selection is appropriate because each observation can change the next research action. The host should retain every tool request and result, continue according to stop_reason, and enforce the deterministic eight-call limit. This combines adaptive reasoning with application-controlled safety, preventing unbounded production loops while avoiding a rigid sequence. Why the other options are wrong: - Option A: A fixed sequence can suit stable workflows, but it cannot adapt when returned evidence indicates a different source or follow-up. - Option B: Discarding results breaks continuity, and the missing host-enforced limit violates the stated production policy. - Option D: Up-front planning can be useful, but the complete tool sequence cannot be known reliably before the agent observes intermediate evidence.

Sample Question 3 — Claude Code Configuration & Workflows

A development team needs Claude Code to run `make verify` before proposing changes to a shared repository. Every contributor must receive this instruction after cloning the repository. One developer also wants Claude Code to use concise explanations in all of her projects. Which configuration is MOST appropriate?

  1. A. Put both instructions in the developer's user-level CLAUDE.md.
  2. B. Put `make verify` in the project CLAUDE.md and the explanation preference in the user's CLAUDE.md. (Correct answer)
  3. C. Put both instructions in the project CLAUDE.md and ask other developers to ignore the style preference.
  4. D. Put `make verify` in a personal slash command and the explanation preference in a directory rule.

Correct answer: B

Explanation: Correct answer (B): The verification command is durable, team-wide repository guidance, so it belongs in the version-controlled project CLAUDE.md. The explanation style is an individual preference spanning projects, so it belongs in the user's CLAUDE.md. Correct configuration scope ensures production workflows are consistently shared without imposing personal preferences on the team. Why the other options are wrong: - Option A: A user-level file is appropriate for the explanation preference, but repository verification instructions stored there would not automatically reach every contributor. - Option C: A project file is appropriate for verification, but committing an individual's presentation preference imposes irrelevant guidance on the entire team. - Option D: A command could invoke an optional workflow, but mandatory verification guidance and a cross-project personal preference do not fit these scopes.

Sample Question 4 — Claude Code Configuration & Workflows

A team frequently asks Claude Code to prepare database migration checklists for a named service and release identifier. The workflow is optional, should be explicitly invoked with those two arguments, and must be versioned for everyone working in the repository. Which implementation is MOST appropriate?

  1. A. Create a project slash command under `.claude/commands/` and document the service and release parameters with an argument hint. (Correct answer)
  2. B. Add the entire checklist workflow to the root CLAUDE.md so it is active during every repository interaction.
  3. C. Create the command under each developer's `~/.claude/commands/` directory and manually keep the copies synchronized.
  4. D. Add a path-specific rule matching all source files so the checklist runs whenever application code is examined.

Correct answer: A

Explanation: Correct answer (A): A project slash command is suitable for an optional, explicitly invoked, repository-shared workflow. An argument hint communicates the required service and release inputs. Matching persistence and sharing scope to invocation semantics avoids consuming context on unrelated work and gives the production team one version-controlled workflow. Why the other options are wrong: - Option B: CLAUDE.md is suitable for durable instructions, but an optional parameterized workflow should not be loaded into every interaction. - Option C: Personal commands are useful for individual workflows, but separate user copies are inappropriate when the team needs one shared version. - Option D: Path rules can conditionally apply file conventions, but examining source code should not automatically trigger an optional release checklist.

Sample Question 5 — Context Management & Reliability

A support agent has accumulated 70,000 tokens of chat history and tool responses while investigating a billing dispute. The agent must retain the exact disputed amount, transaction date, case ID, customer commitments, and unresolved issues. Which architectural change BEST reduces context pressure without risking loss of these facts?

  1. A. Retain every tool response and remove only the customer's older chat messages.
  2. B. Maintain a structured case record, trim processed tool output, and place the current case summary prominently. (Correct answer)
  3. C. Replace the entire history with a short narrative summary generated after each agent turn.
  4. D. Move the oldest messages into the middle of the prompt and preserve recent tool output.

Correct answer: B

Explanation: Correct answer (B): A structured case record preserves exact values and unresolved items while allowing verbose evidence already processed to be removed from active context. The underlying principle is to separate durable, decision-relevant state from transient conversational detail and keep critical state prominent. In production, this reduces token use and lost-in-the-middle failures without turning exact amounts, dates, or identifiers into vague prose. Why the other options are wrong: - Option A: Original tool output preserves evidence, but removing customer messages can discard commitments while retaining large volumes of irrelevant data. - Option C: Narrative summaries reduce context, but they can omit or blur exact amounts, dates, identifiers, and unresolved issues. - Option D: Burying older facts in the middle increases retrieval risk and does not meaningfully control context growth.

Sample Question 6 — Context Management & Reliability

An extraction pipeline processes 400-page insurance files. An upstream agent identifies policy values, and a downstream agent validates them against rules and citations. Passing every source page and the upstream agent's full reasoning exceeds the context budget. Source attribution must be preserved, and the validator must recognize missing fields. Which design is MOST appropriate?

  1. A. Send only a prose summary of the identified policy values and discard the original source pages.
  2. B. Divide the source into equal chunks and let the validator infer relationships without upstream findings.
  3. C. Send the complete source and omit upstream reasoning so all validation uses original evidence directly.
  4. D. Pass a structured fact manifest with values, source locations, and unresolved fields, retaining source text for targeted retrieval. (Correct answer)

Correct answer: D

Explanation: Correct answer (D): A structured manifest makes each extracted value, citation, and unresolved field explicit while allowing the validator to retrieve only the relevant source passages. The underlying principle is that downstream reliability improves when critical state is represented structurally and raw evidence remains available by reference. In production, this controls context size, preserves provenance, and prevents missing information from being mistaken for a confirmed absence. Why the other options are wrong: - Option A: A prose summary is compact, but discarding the source prevents citation verification and can hide unresolved fields. - Option B: Equal-sized chunks can separate related facts and force the validator to reconstruct findings without a reliable manifest. - Option C: Sending all 400 pages still exceeds the stated context budget and makes key fields difficult to locate.

Sample Question 7 — Prompt Engineering & Structured Output

A team uses Claude Code in CI to review pull requests. Developers complain that the reviewer reports harmless formatting preferences, speculative risks, and inconsistent severity labels. The team wants findings limited to defects supported by evidence in the changed code. Which prompt change is MOST appropriate?

  1. A. Ask the reviewer to find every possible issue and let developers dismiss low-value findings.
  2. B. Define report, skip, evidence, and severity criteria, with examples at the decision boundaries. (Correct answer)
  3. C. Ask the reviewer to focus on serious issues while independently deciding what serious means.
  4. D. Increase the response token limit so the reviewer can explain every observation in detail.

Correct answer: B

Explanation: Correct answer (B): Defining explicit report and skip criteria tells the reviewer which observations qualify as findings, while evidence requirements and calibrated severity definitions reduce speculation and inconsistency. The underlying principle is that production prompts should make decision boundaries explicit rather than delegate undefined judgments to the model. This matters in CI because predictable, actionable findings preserve developer trust and reduce review noise. Why the other options are wrong: - Option A: This may maximize recall during exploratory auditing, but it preserves the false-positive burden the team needs to reduce. - Option C: This prioritizes important defects in principle, but the undefined meaning of serious permits inconsistent severity judgments. - Option D: More output space can improve explanations, but it does not define which observations should be reported or skipped.

Sample Question 8 — Prompt Engineering & Structured Output

An insurer uses Claude to extract cancellation dates from customer letters. Some letters quote dates from policy examples or mention a date the customer considered but rejected. The production system must not treat those references as actual cancellation requests, and uncertainty must not be converted into a guessed date. What is the BEST prompt design?

  1. A. Extract every date near the word cancellation and rely entirely on downstream date validation.
  2. B. Return the earliest valid date because customers usually discuss events in chronological order.
  3. C. Require one cancellation date in every response and ask Claude to infer it from context.
  4. D. Define positive, negative, and ambiguous cases, and permit an explicit uncertain or missing result. (Correct answer)

Correct answer: D

Explanation: Correct answer (D): Positive criteria identify genuine cancellation requests, negative criteria exclude examples and rejected dates, and ambiguous criteria allow uncertainty to be represented without fabrication. The underlying principle is that precise prompt boundaries complement downstream validation by controlling semantic interpretation. This matters in production because a syntactically valid but unsupported cancellation date could trigger an incorrect policy action. Why the other options are wrong: - Option A: Proximity-based extraction is simple and favors recall, but date-format validation cannot determine whether a date was illustrative or rejected. - Option B: A chronological heuristic may help in some document types, but it is unsupported here and can select an unrelated date. - Option C: Requiring a value simplifies downstream handling, but forcing inference violates the requirement not to guess when evidence is uncertain.

Sample Question 9 — Tool Design & MCP Integration

A customer-support agent has three tools named customer_lookup, account_lookup, and profile_lookup. Their descriptions all say, "Find customer information," but each tool queries a different system and accepts a different identifier. Production traces show that Claude frequently selects the wrong tool. What is the BEST architectural change?

  1. A. Combine the tools into one interface with optional parameters for every supported identifier and backend.
  2. B. Rename and document each tool with its purpose, accepted identifier, output, boundaries, and selection examples. (Correct answer)
  3. C. Keep the interfaces unchanged and instruct Claude to retry another tool whenever no customer is returned.
  4. D. Provide all three tools through separate MCP servers while retaining their current names and descriptions.

Correct answer: B

Explanation: Correct answer (B): The tools need distinct semantic boundaries and descriptions that state their purpose, inputs, outputs, and appropriate selection conditions. Examples can further clarify similar cases. Clear interfaces reduce model selection ambiguity, which matters in production because choosing the wrong customer system can cause incorrect conclusions or actions. Why the other options are wrong: - Option A: Combining distinct backends and identifier types behind many optional parameters would preserve or increase ambiguity rather than create clear boundaries. - Option C: An empty result can be valid, so speculative retries against unrelated systems are unsafe and do not resolve the underlying interface ambiguity. - Option D: Separate servers may support security or ownership boundaries, but server separation alone does not clarify ambiguous tool names and descriptions.

Sample Question 10 — Tool Design & MCP Integration

A research agent uses a tool named research_data with a mode parameter supporting patent search, market-metric retrieval, document summarization, and citation export. The modes have different required inputs, permissions, outputs, and failure conditions. Agents frequently omit mode-specific fields and misinterpret results. Which redesign is MOST appropriate?

  1. A. Split the interface into purpose-specific tools with distinct schemas, boundaries, outputs, and representative usage examples. (Correct answer)
  2. B. Retain one tool, make every field optional, and have the server infer the intended mode from supplied values.
  3. C. Retain one tool and expand its description with a complete list of every backend implementation detail.
  4. D. Create one tool per input field so agents can construct each research operation over multiple calls.

Correct answer: A

Explanation: Correct answer (A): The four modes represent materially different operations with distinct schemas, permissions, outputs, and failure behavior, so purpose-specific tools provide coherent boundaries. This follows the principle that tools should expose understandable capabilities rather than unrelated modes. In production, clearer schemas improve selection, validation, authorization, and result interpretation. Why the other options are wrong: - Option B: Making fields optional and inferring the mode preserves ambiguity and makes malformed invocations more likely. - Option C: Additional description can help a cohesive tool, but backend implementation details do not resolve incompatible schemas and operational boundaries. - Option D: Dividing operations by individual input fields creates excessive fragmentation and shifts unnecessary assembly work to the agent.

Quick 10-Question CCAR-F Practice Test

Take a free 10-question CCAR-F quick-start practice test covering all 5 exam domains. Get instant scoring with detailed explanations — perfect for a quick readiness check.

About the Claude Certified Architect – Foundations Certification

CCAR-F is part of Anthropic's certification program, designed to validate the agentic AI skills employers increasingly need in 2026. It sits at the architect level: rather than simply calling the Claude API, certified architects design agentic systems, choose orchestration patterns, plan MCP integration architecture, and build for production reliability. Anthropic recommends hands-on experience building with Claude, familiarity with the Claude Agent SDK, MCP, and Claude Code, plus the free courses on Anthropic Academy. The certification is valid for 12 months, and because Anthropic's program is evolving quickly alongside the Claude platform, expect updated exam content at renewal.

CCAR-F vs CCDV-F: Which Anthropic Certification Should You Take?

Anthropic offers two foundations-level credentials that target different roles in the Claude ecosystem:

CriteriaCCAR-FCCDV-F
CertificationClaude Certified Architect – FoundationsClaude Certified Developer – Foundations
LevelArchitect (system design)Developer (application building)
Core FocusAgentic system design, MCP architecture, production reliabilityClaude API, tool use, application patterns
Heaviest DomainAgentic Architecture & Orchestration (27%)Claude API and tool use
PrerequisiteNoneNone
Exam Cost (US)$125$125
Validity12 months12 months

Many candidates take CCDV-F first to build API fluency, but neither is a prerequisite for the other. If you already design agentic systems and reason about orchestration patterns, MCP integration, and production reliability, CCAR-F is the credential that matches your work. Deciding between the Claude and cloud paths? Read Claude Architect vs AWS AI Practitioner to see which AI certification fits your 2026 career goals.

CCAR-F Study Plan — 2 to 4 Weeks

Week 1 — Foundations: Read the official CCAR-F exam objectives and work through the free Anthropic Academy courses. Set up Claude Code, connect an MCP server, and build a small agent with the Claude Agent SDK. Take a 10-question diagnostic to find weak domains.

Weeks 2–3 — Domain drilling: Prioritize the heaviest domain — Agentic Architecture & Orchestration (27%) — then Claude Code Configuration & Workflows and Prompt Engineering & Structured Output (20% each). Practice CCAR-F questions domain by domain, reviewing every wrong answer and targeting 75%+ per domain. Get hands-on with tool schemas, MCP transports, prompt caching, and compaction.

Week 4 — Timed mocks: Take 2–3 full-length timed practice tests. When you consistently score 75%+ across all five domains, register and sit the exam. Candidates with hands-on Claude experience typically need 2–4 weeks; allow more time if you haven't used MCP or Claude Code in production.

CCAR-F Salary and Career Outlook

AI and agent engineers typically earn $120K–$180K USD in the US, with hands-on Claude and agentic-system experience commanding premiums at the top of that range. As one of the first certifications covering agentic architecture and MCP integration, CCAR-F signals exactly the skill set employers are hiring for in 2026 — demand for engineers who can design and ship production agent systems continues to outpace supply.

Frequently Asked Questions

What is the Claude Certified Architect – Foundations (CCAR-F) exam?

CCAR-F is Anthropic's certification for practitioners who design and build production systems with Claude. It validates your ability to architect agentic systems, design tools and MCP integrations, configure Claude Code workflows, engineer prompts with structured output, and manage context and reliability in production.

How much does the CCAR-F exam cost?

The CCAR-F exam costs $125 USD. Anthropic occasionally offers vouchers and discounts through Anthropic Academy programs and partner training bundles.

What score do I need to pass CCAR-F?

You need a scaled score of 720 or higher on Anthropic's 100–1,000 scale. This is a scaled score, not a percentage — the exact number of questions you must answer correctly varies with question difficulty.

How many questions are on the CCAR-F exam and how long is it?

The CCAR-F exam has 60 questions and you get 120 minutes — an average of two minutes per question. Questions are scenario-based multiple choice, testing practical architectural judgment rather than pure recall.

What are the CCAR-F exam domains and their weights?

Five domains: Agentic Architecture & Orchestration (27%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%), Tool Design & MCP Integration (18%), and Context Management & Reliability (15%). Agentic architecture carries the most weight.

What is the difference between CCAR-F and CCDV-F?

CCDV-F (Claude Certified Developer – Foundations) focuses on building with the Claude API as a developer — requests, tool use, and application patterns. CCAR-F (Architect – Foundations) sits a level up: designing agentic systems, choosing orchestration patterns, MCP integration architecture, and production reliability. Many candidates take CCDV-F first, but neither is a prerequisite for the other.

Do I need coding experience for CCAR-F?

Yes — you should be comfortable reading and reasoning about code that uses the Claude API, the Claude Agent SDK, and MCP servers. The exam emphasizes architectural decisions (which pattern, which configuration, what to do first) rather than writing code from scratch, but hands-on experience with Claude Code and the API is essential.

Are there prerequisites for the CCAR-F exam?

There are no formal prerequisites — you can register and sit the exam directly. In practice, Anthropic recommends hands-on experience building with Claude, familiarity with the Claude Agent SDK, MCP, and Claude Code, plus the free courses on Anthropic Academy.

How hard is the CCAR-F exam?

CCAR-F is an intermediate exam focused on applied judgment. Candidates find Agentic Architecture & Orchestration the most demanding domain because it requires choosing between orchestration patterns (workflows vs agents, orchestrator–worker, subagent delegation) under realistic constraints. Scenario-based practice questions are the most reliable preparation.

How long should I study for CCAR-F?

Most candidates with hands-on Claude experience need 2–4 weeks of preparation. Plan more time if you haven't used MCP or Claude Code in production. Prioritize the heaviest domain — Agentic Architecture & Orchestration (27%) — then Claude Code workflows and prompt engineering (20% each).

How long is the CCAR-F certification valid?

The Claude Certified Architect – Foundations certification is valid for 12 months. Anthropic's certification program is evolving quickly alongside the Claude platform, so expect updated exam content at renewal.

Is it CCA-F, CCAF, or CCAR-F?

All three refer to the same certification — Anthropic's Claude Certified Architect – Foundations. The official exam code is CCAR-F, but the exam is widely abbreviated as CCA-F or CCAF in study guides and community discussions. Whichever abbreviation you searched for, this is the practice test for that exam.

Are these CCAR-F practice questions free?

Yes — the sample questions on this page are free with no registration, and every FlashGenius account includes 10 free practice questions per day from the full 250+ CCAR-F question bank, each with four answer options, the correct answer, and a detailed explanation. Premium ($14.99/month) unlocks unlimited access to the entire bank plus exam simulation and smart review — a fraction of the $125 exam fee and cheaper than most paid CCA-F practice platforms.

Go Deeper: CCAR-F Practice Questions and Guides

Official Anthropic CCAR-F Resources

Pair your practice questions with Anthropic's official preparation materials:

Start your free CCAR-F practice test now | All Sample Tests

Related Certifications

AWS Certified AI Practitioner Practice Test | Microsoft AI-103 Practice Test | AWS ML Engineer Associate Practice Test | Databricks Generative AI Engineer Practice Test