The Claude Certified Architect – Foundations certification (exam code CCAR-F, often written CCA-F) is Anthropic's credential for people who design and build production systems with Claude. Anthropic describes it as validating that practitioners can make informed decisions about trade-offs when implementing real-world solutions, across Claude Code, the Claude Agent SDK, the Claude API and the Model Context Protocol (MCP). It launched in March 2026 alongside the Claude Partner Network.
Format and scoring
| Question types | Multiple choice and scenario-based multiple response |
|---|---|
| Number of questions | 60, as reported in exam-guide summaries (the official FAQ does not state a count) |
| Exam time | 120 minutes; plan for about 135 minutes of seat time |
| Score | Scaled 100–1,000; pass mark 720 |
| Delivery | Pearson online proctoring (OnVUE) or a Pearson test center |
| Rules | Closed book: no notes, documentation, AI assistants or translation tools |
| Language | English, according to the official FAQ |
Your score appears on screen at the end. The score report shows the overall scaled score, pass or fail, and the percentage correct in each section, and passing candidates are invited to claim a Credly badge. Because scaling is not published, a raw percentage cannot be converted exactly into the 100–1,000 score.
The five domains
The exam guide lists five weighted domains. Agentic architecture is the largest block; context management is the smallest but is widely reported as the hardest because its answers depend on precise mechanics.
| Domain | Weight | ≈ Questions of 60 |
|---|---|---|
| 1. Agentic Architecture & Orchestration | 27% | 16 |
| 2. Claude Code Configuration & Workflows | 20% | 12 |
| 3. Prompt Engineering & Structured Output | 20% | 12 |
| 4. Tool Design & MCP Integration | 18% | 11 |
| 5. Context Management & Reliability | 15% | 9 |
Scenario-based questions
Questions are set inside realistic situations. Six scenario themes are reported, with four drawn at random for each sitting: 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. A scenario tests several domains at once, so the same support-agent setting might ask about loop design, tool errors and caching. See our scenario guide for what each one emphasises.
What the questions reward
The exam does not ask you to recall API trivia. It gives you a running system with a problem and several plausible fixes. The winning answer usually puts the control at the right layer: deterministic code or permissions for rules that must hold, better interfaces for tool problems, measurement for prompt changes, and the simplest architecture that meets the requirement. Distractors tend to be prompt-only fixes for hard rules, irrelevant knobs like temperature, or rebuilds that are far larger than the problem.
Who can take it
Registration is currently limited to employees of Claude Partner Network organizations using a recognised company email; personal addresses and non-partners are not eligible for now. There is no prerequisite certification. Fees, retakes and renewal are covered on the cost and registration page.
How to prepare
Read the official exam guide first: it is the authoritative list of domains and task statements and includes sample questions. Anthropic retired its practice exam when delivery moved to Pearson, so most practice now comes from building things and from independent question banks like ours. Start with the 10-question diagnostic, follow the four-week plan, and finish with a timed 60-question practice exam to rehearse the two-minute-per-question pace.
Facts on this page were checked on November 22, 2026 against Anthropic's certification FAQ. Exam rules can change; the official exam guide and FAQ are authoritative.