MockExam Space

CCA-F practice exam: free Claude Certified Architect mock test

Prepare for Anthropic's Claude Certified Architect - Foundations exam with mocks that mirror the real format: 60 questions, 120 minutes, five weighted domains, and a 72% passing bar. Start with the 10 sample questions below, two from each domain.

The five CCA-F domains

Agentic Architecture & Orchestration

Agent SDK, subagents, orchestration patterns

Tool Design & MCP Integration

Tool schemas, MCP primitives and transports

Claude Code Configuration & Workflows

Settings precedence, CLAUDE.md, hooks, skills

Prompt Engineering & Structured Output

Prompting method, XML tags, structured output

Context Management & Reliability

Context windows, compaction, caching, evals

Sample CCA-F questions

Scenario-style questions like the real exam, with an explanation for every answer.

Question 1 路 Agentic Architecture & Orchestration

You are scaffolding a new TypeScript agent and need the single call that starts the agentic loop and streams messages back as an async iterator. Which Agent SDK entry point do you reach for?

  1. A.run(), then poll a status endpoint for messages
  2. B.createAgent(), which returns an agent object you invoke
  3. C.invoke(), passing a callback for each message
  4. D.query(), and iterate the messages it yields
Show answer and rationale

Correct answer: D

query() is the SDK entry point in both TypeScript and Python; it starts the agentic loop and yields messages as an async iterator. The other names do not exist in the SDK.

Question 2 路 Agentic Architecture & Orchestration

A teammate is adding the Agent SDK to both a Node service and a Python data pipeline and asks which packages to install. Which pair is correct?

  1. A.@anthropic-ai/claude-code and claude-code-sdk
  2. B.@anthropic/agent-sdk and anthropic-agent-sdk
  3. C.@anthropic-ai/claude-agent-sdk (Node 18+) and claude-agent-sdk (Python 3.10+)
  4. D.@claude/agent-sdk and pyclaude-agent
Show answer and rationale

Correct answer: C

TypeScript uses @anthropic-ai/claude-agent-sdk (Node 18+); Python uses claude-agent-sdk (Python 3.10+). The claude-code packages are the older Claude Code SDK naming.

Question 3 路 Tool Design & MCP Integration

You are building an MCP server and planning what to expose. What are the three core primitives an MCP server can provide?

  1. A.Functions, webhooks, and schemas
  2. B.Tools, resources, and prompts
  3. C.Actions, documents, and templates
  4. D.Endpoints, models, and contexts
Show answer and rationale

Correct answer: B

MCP servers expose tools (model-invoked actions), resources (readable data), and prompts (user-invoked templates).

Question 4 路 Tool Design & MCP Integration

You will ship one MCP server that runs as a local process and another hosted remotely. Which two transports are the standard choices?

  1. A.stdio for the local process and streamable HTTP for the remote server
  2. B.gRPC for local and GraphQL for remote
  3. C.WebSocket only, with polling as a fallback
  4. D.Unix sockets locally and FTP remotely
Show answer and rationale

Correct answer: A

Local servers typically run over stdio; remote servers use streamable HTTP (which superseded SSE).

Question 5 路 Claude Code Configuration & Workflows

A permission rule in a user's personal settings conflicts with an enterprise managed policy in a regulated org. Which one takes effect?

  1. A.User ~/.claude/settings.json always wins
  2. B.Enterprise managed policy settings override project and user settings
  3. C.The most recently modified file wins regardless of scope
  4. D.Project settings.json and settings.local.json merge randomly
Show answer and rationale

Correct answer: B

Precedence runs enterprise managed policies > command-line args > project local > project shared > user settings. Managed policy is designed to be non-overridable.

Question 6 路 Claude Code Configuration & Workflows

A team keeps re-explaining the same conventions to Claude Code every session. What is a CLAUDE.md at the repo root for?

  1. A.It is automatically pulled into context to give persistent project instructions and conventions
  2. B.It configures which model tier the CLI may use
  3. C.It stores encrypted credentials for MCP servers
  4. D.It is only read when the user runs /init explicitly
Show answer and rationale

Correct answer: A

CLAUDE.md holds project instructions and conventions loaded into context each session. /init can generate one, but it is read every session, not only on demand.

Question 7 路 Prompt Engineering & Structured Output

A team wants to 'start prompt engineering' a new feature immediately. Per Anthropic's guidance, what should already exist before they begin iterating on prompts?

  1. A.A fine-tuned model checkpoint for the task
  2. B.Success criteria, a way to test empirically, and a first-draft prompt
  3. C.A complete library of few-shot examples for every edge case
  4. D.An approved production deployment plan
Show answer and rationale

Correct answer: B

The prompt engineering process assumes defined success criteria, an empirical way to test against them, and a first draft to iterate on. Without those, iteration is unmeasured guessing.

Question 8 路 Prompt Engineering & Structured Output

Your app meets accuracy targets but latency is far too high for production. Per Anthropic guidance, what is the most appropriate first consideration?

  1. A.Whether a different (faster) model choice fits the latency/cost constraint
  2. B.Adding more chain-of-thought to make answers more efficient
  3. C.Increasing max_tokens so responses complete in fewer turns
  4. D.Rewriting the prompt in XML to speed up parsing
Show answer and rationale

Correct answer: A

Not every failing metric is a prompt problem; latency and cost issues may warrant selecting a faster or smaller model. More chain-of-thought would increase latency, not reduce it.

Question 9 路 Context Management & Reliability

A long-running agent session keeps growing and you have added no mitigation. What happens as it approaches the model's context window limit?

  1. A.The API transparently swaps to a bigger model
  2. B.Older messages remain and new output quality is unaffected
  3. C.Requests eventually fail or earlier context must be dropped, losing information the agent may need
  4. D.Tokens are compressed automatically at no fidelity cost
Show answer and rationale

Correct answer: C

The context window is finite; without compaction or summarization, long sessions hit the limit and either fail or drop earlier context, losing information.

Question 10 路 Context Management & Reliability

Your agent must keep working well past what fits in one context window. What is the purpose of compaction/summarization?

  1. A.To replace older conversation history with a summary so work can continue within the limit
  2. B.To reduce billing by deleting cached prefixes
  3. C.To encrypt the transcript before persistence
  4. D.To merge parallel subagent transcripts into one file
Show answer and rationale

Correct answer: A

Compaction summarizes earlier context and carries the summary forward, letting the session continue past the raw context limit.

Frequently asked questions

What is the CCA-F certification?

CCA-F (Claude Certified Architect - Foundations) is Anthropic's foundational certification for building with Claude: the Agent SDK, MCP, Claude Code, prompt engineering, and context management. It is the entry point of the Claude Certified Architect track.

What is the CCA-F exam format and passing score?

The exam is 60 questions in 120 minutes across five weighted domains, with a passing bar of 72 percent. Our mock exams use the same question count, time limit, domain weighting, and pass threshold.

How should I prepare for the CCA-F?

Combine hands-on time with Claude Code and the Agent SDK with targeted practice questions. Our bank has 1,000+ scenario questions plus five 60-question model sets; every question explains why the correct answer is right, and the AI tutor can go deeper on any topic.

Is this practice exam free?

The 10 sample questions on this page need no account. During launch, a free account unlocks all model sets, timed simulation, per-domain score breakdowns, and the AI tutor.

1,000+ scenario questions and five full model sets

Timed 60-question simulations with per-domain scoring against the 72% bar, plus an AI tutor for follow-up questions. Free during launch.