Anthropic · CCAR-F
Claude Certified Architect – Foundations (CCAR-F) Practice Tests
Validates foundational knowledge and skills for designing and operating production-grade agentic AI systems using Claude, including the Claude API, Agent SDK, Claude Code, and Model Context Protocol (MCP). Targeted at software engineers, solutions architects, and AI developers building real-world Claude-powered applications.
Exam Details
Practice Questions
314
≈ 4 practice exams
Duration
120 minutes
Passing Score
720/1000
Difficulty
FoundationalLast Updated
Sep 2026
Topics Covered
Agentic Architecture & OrchestrationClaude Code Configuration & WorkflowsTool Design & MCP IntegrationPrompt EngineeringContext Management
Claude Certified Architect – Foundations (CCAR-F) Practice Exam Preparation
Use this CCAR-F practice exam to prepare for Claude Certified Architect – Foundations (CCAR-F) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 314 questions for Anthropic CCAR-F, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Agentic Architecture & Orchestration, Claude Code Configuration & Workflows, Tool Design & MCP Integration, Prompt Engineering, and Context Management. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are written for candidates preparing for Anthropic certification questions and reviewing how the concepts apply in practice. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Sample Questions
5 sample questions with answers and explanations. The full bank has 314 questions, enough for 4 full-length practice exams.
Preview — answers shown1. An architect is designing a shared CLAUDE.md for a team. They want to include a common set of coding standards by importing a file from a shared location, but they are concerned about the risk of circular imports and excessive nesting. What is the maximum recursion depth allowed for @import statements in Claude Code? (Select one!)
AFive hops
BTwo hops
CThree hops
DFour hops
Explanation
The author's key is correct. According to the official Claude Code documentation on memory, CLAUDE.md files can import additional files using the `@path/to/import` syntax, but this feature is limited to a maximum recursion depth of four hops. This constraint is explicitly designed to prevent circular imports and excessive nesting, which could otherwise lead to context bloat or infinite loops. Exceeding this limit causes the import to fail silently or be ignored, so architects must structure their import hierarchy to stay within four levels.
2. You are using the Claude API with the default tool_choice 'auto' and a set of tools. You want to estimate the token cost of the tool-use system prompt. According to the official documentation, what is the token cost for the tool-use system prompt when using Claude Sonnet 4.6 with tool_choice 'auto'? (Select one!)
A290 tokens
B675 tokens
C497 tokens
D589 tokens
Explanation
The official documentation provides a table of tool-use system prompt token costs by model and tool_choice. For Claude Sonnet 4.6 with tool_choice 'auto' or 'none', the cost is 497 tokens. The 290-token figure is for Opus 4.8 with 'auto'/'none'. The 589-token figure is for Sonnet 4.6 with 'any'/'tool'. The 675-token figure is for Opus 4.7 with 'auto'/'none'. Therefore, 497 tokens is the correct answer for Sonnet 4.6 with 'auto'.
3. A team is building a multi-agent system for a customer support workflow. The workflow involves three sequential stages: first, classify the customer's issue; second, retrieve relevant knowledge base articles; third, draft a response. Each stage depends on the output of the previous stage, and the stages are always executed in the same order. The team wants to choose an orchestration pattern that matches this predictable, fixed workflow. Which pattern is most appropriate? (Select one!)
APrompt chaining, where each stage is a separate step in a fixed sequential pipeline
BA single agent with all tools and a detailed system prompt
CDynamic adaptive decomposition, where the plan evolves as the agent discovers new information
DHub-and-spoke with a coordinator that dynamically decides which subagent to invoke next
Explanation
Prompt chaining is the correct pattern for a fixed, predictable sequence of stages where each step depends on the previous one. This matches the described workflow of classify, retrieve, and draft. Hub-and-spoke with a coordinator is for tasks that require dynamic delegation and aggregation, not a fixed pipeline. Dynamic adaptive decomposition is for open-ended investigation where the plan changes as new information is found. A single agent with all tools would be simpler but would not provide the structured separation of stages that prompt chaining offers for this predictable multi-aspect review.
4. You are building a multi-agent research system with a coordinator and several subagents. The coordinator needs to delegate three independent research tasks to three subagents. You want all three subagents to run in parallel to minimize latency. How should the coordinator invoke the subagents? (Select one!)
AEmit three separate Agent tool calls across three consecutive turns, waiting for each to complete.
BUse a single Agent tool call that takes a list of tasks as input.
CEmit three Agent tool calls in a single assistant response, allowing parallel execution.
DUse a background agent for each task and then collect results via cross-session messaging.
Explanation
To spawn parallel subagents, the coordinator must emit multiple Agent (or Task) tool calls in a single assistant response. This allows the SDK to execute them concurrently. Emitting them across separate turns serializes execution. There is no single Agent tool call that takes a list of tasks; each subagent requires its own tool call. Background agents are for independent sessions, not for parallel subagents within the same session.
5. You are building an agent that uses a mix of Anthropic-provided tools. You need to understand the three categories of these tools to configure them correctly. Which two statements correctly describe the categories? (Select two!)
Multiple correct answersAUser-defined client tools are defined by you in the top-level tools parameter.
BAnthropic-schema client tools have schemas that are trained-in, so Claude calls them more reliably.
CServer-executed tools run on your own infrastructure and are billed like any API request.
Dmcp_toolset is a user-defined client tool that connects to MCP servers.
EAnthropic-schema client tools are executed on Anthropic's servers.
Explanation
User-defined client tools are indeed defined by you in the top-level tools parameter. Anthropic-schema client tools have schemas that are trained-in, making Claude call them more reliably. Server-executed tools run on Anthropic's servers, not your own infrastructure, and have usage-based pricing. mcp_toolset is a server-executed tool, not a user-defined client tool. Anthropic-schema client tools are executed on the client side, not on Anthropic's servers.
$17.99
One-time access to this exam
314 questions (4 practice exams' worth)
Unlimited timed exam simulations
Yours forever — no subscription
Detailed explanations
Free preview stays available