The CCA-F (Certified Claude Associate — Foundations) is a 60-question, scenario-based exam designed for engineers who build production systems with Claude. This guide covers the exam structure, scoring, and what you can expect from each domain.
Exam at a glance
| Questions | 60 |
| Duration | 90 minutes |
| Score range | 100–1000 |
| Passing score | 720 |
| Format | Scenario-based multiple choice |
| Credential validity | 6 months |
Exam domains
The CCA-F is divided into five domains, each weighted by its importance in real-world production engineering:
1. Agentic Architecture & Orchestration (27% — 16 questions)
The largest domain. Tests your ability to design multi-agent systems using Claude: how to structure subagent delegation, handle agent failures gracefully, and maintain coherent task state across long-running workflows. Expect scenarios that require you to choose between parallelization strategies, identify when tool results should feed back into prompt context, and recognize failure modes in poorly-scoped agents.
2. Tool Design & MCP (23% — 14 questions)
Tool use and the Model Context Protocol. How to write tool definitions Claude can reliably interpret, when to use MCP vs. direct API calls, and how to structure tool results for effective downstream reasoning. Scenarios often involve diagnosing why an agent is misusing a tool and correcting the schema or description.
3. Prompt Engineering & Structured Output (18% — 11 questions)
Not theoretical — applied. Scenarios test whether you can diagnose over-specified prompts, extract reliable structured data from Claude, and write system prompts that behave consistently across edge cases. The emphasis is on production reliability, not prompt writing as a craft.
4. Claude Code for Production (17% — 10 questions)
Working with Claude Code in CI/CD pipelines, multi-file refactors, and automated code review workflows. How to configure Claude Code effectively, when to use hooks, and how to scope large tasks without losing context coherence.
5. Context Management & Reliability (15% — 9 questions)
Context window management, caching strategies, and reliability patterns for high-traffic production deployments. Scenarios include diagnosing context overflow in long conversations and choosing between summarization, compression, and windowing approaches.
Question format
All questions are scenario-based. You're given a situation — a broken agent, a poorly performing tool, an architecture decision to make — and four options. The wrong answers are plausible; they reflect common engineering mistakes, not obviously bad choices. This is the core difference from trivia-based AI exams.
Example scenario structure:
An orchestrator agent delegates research tasks to three subagents in parallel. Subagent B returns a result that contradicts Subagent A. The orchestrator needs to resolve the conflict and continue. Which pattern best handles this without re-running both subagents from scratch?
Scoring
The 100–1000 scale is a scaled score, not raw percentage. A 720 passing score corresponds to approximately 70–75% raw accuracy depending on question difficulty distribution in your exam sitting. Your score report breaks down performance by domain, so you know exactly where to focus if you need to re-sit.
How to practice
Certify's practice sessions use the same 4-scenario, scenario-based format as the real exam. Free users get 5 questions per scenario (preview format). Pro users get full 15-question scenarios — the same length as the actual exam.
Each practice sitting generates a domain breakdown showing your relative strength and weakness areas. The most common weak domain for first-time candidates is Agentic Architecture — specifically multi-agent failure handling.