Claude Certified Architect — Foundations (CCAR-F): complete study guide

The CCAR-F does not test whether you can use Claude. It tests whether you can architect a production system on Claude — and whether you know when to stop trusting a prompt and reach for code.

What you are actually being tested on

Format, scoring, and how the questions are built

Every question is anchored to a production scenario — a support agent, a research system, a CI pipeline — and asks you to choose an architecture, not recall a fact. Four options; exactly one correct. The distractors are written by people who know the material, so they are architecturally plausible. Most of them fail on a single trade-off.

Scoring is scaled (720 of 1000 to pass), so a raw percentage is not a reliable predictor. Treat 80%+ on a realistic practice set as your comfort threshold. Four attempts are permitted per rolling 12 months, with 14-, 30-, and 90-day waits between successive attempts, so the retake path is slow — prepare to pass on the first sitting.

Domain weights

Domains 1 and 3 together are 47% of the exam. If your study time is short, spend it there — but note that Domain 5's reliability thinking bleeds into every other domain's questions, so it is worth more than its 15% suggests.

How to read a question — Find the constraint. Nearly every stem contains one hard requirement — "must never process a refund before verification", "must not add latency", "must produce citations" — and the correct answer is the only option that enforces it rather than requesting it. Options that enforce with code beat options that request with language, every time.

Every page in this guide

• The mental model — The single principle behind most CCAR-F answers: deterministic mechanisms beat probabilistic ones, with a full wrong-instinct versus exam-answer table.
• Domain 1 — Agentic Architecture & Orchestration — The largest CCAR-F domain: the agent loop and stop_reason, subagents and context isolation, AgentDefinition fields, parallel versus sequential, and task-scoped tools.
• Domain 2 — Tool Design & MCP Integration — Tool schemas, tool_choice, parallel tool use, structured errors with isRetryable, MCP primitives and transports, and .mcp.json scopes in Claude Code.
• Domain 3 — Claude Code Configuration & Workflows — CLAUDE.md hierarchy and loading order, settings precedence, permission modes and rule syntax, hooks and exit codes, skills, and headless mode for CI/CD.
• Domain 4 — Prompt Engineering & Structured Output — Prompting techniques in the order to apply them, JSON schema outputs versus strict tool use, schema limits, and validated extraction with error-specific retries.
• Domain 5 — Context Management & Reliability — Context editing parameters, what survives /compact, prompt caching breakpoints and pricing, the Batches API, and which HTTP errors are retryable.
• The 14 anti-patterns — Fourteen distractors that recur across CCAR-F scenarios, each with why it fails and what the exam wants instead. Learn the shape and eliminate options on sight.
• The six exam scenarios — Four of six scenarios appear on any exam form. What each one tests and the traps built into it, from the support agent to structured data extraction.
• Cheat sheet — Stop reasons, settings and MCP precedence, cache pricing, batch limits, the config file map, and which mechanism to use for which requirement.
• Questions & answers — Domain 1 — Sixteen CCAR-F practice questions on agentic architecture and orchestration, each with the correct answer and a full explanation of why the distractors fail.
• Questions & answers — Domain 2 — CCAR-F practice questions on tool schemas, structured errors, MCP primitives, transports and .mcp.json scopes, with answers and full explanations.
• Questions & answers — Domain 3 — CCAR-F practice questions on CLAUDE.md hierarchy, settings precedence, permission rules, hooks and headless CI runs, with answers and full explanations.
• Questions & answers — Domain 4 — CCAR-F practice questions on structured outputs, strict tool use, schema limits and validated extraction, with answers and full explanations.
• Questions & answers — Domain 5 — CCAR-F practice questions on context editing, compaction, prompt caching, the Batches API and retry policy, with answers and full explanations.
• Facts to memorise — Thirty recall facts covering stop reasons, precedence orders, hook exit codes, cache pricing and batch limits for the CCAR-F exam.
• Six-week study plan — A week-by-week CCAR-F preparation plan, from hand-writing a tool-use loop to drilling 100 practice questions in your weakest domain.
• Official resources — Anthropic Academy, Agent SDK and Claude Code docs, the MCP specification, structured outputs and Pearson VUE scheduling, mapped to the exam domains.

This is an independent study guide, not an official Anthropic publication. Exam blueprints change; confirm the current domain weights and format against the official exam guide before you sit.

Next: The mental model →