What the Claude Certified Architect - Foundations exam tests
By Stephen Cheng · Updated 14 Nov 2026 · First published on dev.to, 22 Jun 2026
Anthropic introduced the Claude Certified Architect certification, and the Foundations exam is less about trivia and more about judgment: given a realistic agent-engineering situation, what is the right call? I could not find a good free way to practise that, so my team at CyberSkill built one. This guide breaks down what the exam tests and how we think about each area.
Anthropic's code for the exam is CCAR-F. You will also see it written as CCA-F or CCAF.
The shape of the exam
According to Anthropic's exam guide (version 1.0, July 2026), the exam has 60 multiple-choice and multiple-response questions and a time limit of 120 minutes. The passing score is 720 on a scale of 100 to 1,000. The fee is $125 and the credential is valid for 12 months.
Every question sits inside a scenario. A sitting presents 4 scenarios, picked at random from these 6:
- Customer support resolution agent
- Code generation with Claude Code
- Multi-agent research system
- Developer productivity with Claude
- Claude Code for continuous integration
- Structured data extraction
The questions are scored against five domains:
| Domain | Share of the exam |
|---|---|
| Agentic architecture and orchestration | 27% |
| Tool design and MCP integration | 18% |
| Claude Code configuration and workflows | 20% |
| Prompt engineering and structured output | 20% |
| Context management and reliability | 15% |
Our mock has the same length and time limit, a score at the end and an explanation of every option. It is built on four of the six scenarios, described below. The two Claude Code scenarios are not in it yet. Exam details change, so check the guide before you book; the exam page lists the facts we hold and when we last checked them.
1. Research pipelines
Multi-agent orchestration, recovering state after a failure, and handing context between a coordinator and its sub-agents without bloating the window.
The recurring lesson: persist compact, structured results per agent and resume only the unfinished work, rather than replaying full histories or starting over.
2. Extraction pipelines
Tool contract design and reliable structured output when inputs are ambiguous.
The dependable pattern is to make extraction a contract: a strict schema or tool that the model must satisfy, plus validation and a retry, instead of hoping a prompt returns clean JSON.
3. Customer support agents
Graceful degradation and honest failure handling.
When a tool is down, the right behaviour is to say so plainly, give a realistic next step, and offer to escalate, not to guess a status or give a false assurance.
4. Code exploration
Working in large codebases and scoping an agent's search.
The context-efficient move is to search for the known identifier first and open only the files that match, rather than reading everything into context.
Try it
The sample questions come with their answers and need no account, if you want a taste before a full sitting.
- Learn mode checks each answer as you go and shows the explanation straight away.
- A practice set is untimed and graded when you submit, the way the real exam works.
- The timed mock runs all 60 questions against the clock.
All three start from the exam page. The practice is unofficial and not affiliated with Anthropic. If you take it, I would value your read on whether the questions match the real exam's difficulty: tell us on the Status Hub.
Practise Claude Certified Architect - Foundations
Every question is free. Start with the sample questions, or go straight to the exam page.
These questions are free, with no ads. Practice depends on learners who choose to support it. How support works