Claude Certified Architect – Foundations
For solution architects building production apps with the Claude API, the Agent SDK, Claude Code and MCP.
The real exam
- Questions
- 60
- Time
- 120 minutes
- Format
- Scenario-based multiple choice
- Scoring
- Scaled score; 720 out of 1,000 is widely reported as the pass mark
- Exam fee
- About US$125 (reported)
- Practice pass mark
- 72%
Exam facts come from third-party reports as of November 2026, because Anthropic doesn't publish every detail on public pages. Domain weights are reported, not official. Always confirm on the official exam page before you book.
Practice
Practice exams
Fixed sets with no questions in common. Take one as a timed exam, or in review mode to see explanations after each question.
-
Diagnostic set
6 questions, about 12 minutes. A short mix across every domain to find your weak spots.
-
Practice exam 1
60 questions, about 120 minutes. Full length, in the real exam's domain mix
-
Practice exam 2
60 questions, about 120 minutes. Full length, in the real exam's domain mix
-
Practice exam 3
60 questions, about 120 minutes. Full length, in the real exam's domain mix
-
Practice exam 4
60 questions, about 120 minutes. Full length, in the real exam's domain mix
-
Practice exam 5
60 questions, about 120 minutes. Full length, in the real exam's domain mix
-
Practice exam 6
60 questions, about 120 minutes. Full length, in the real exam's domain mix
Randomized test
Questions drawn from the whole bank in the real exam's domain proportions, in a new order each time.
-
Full length
60 questions, about 120 minutes
-
Short
20 questions, about 40 minutes
-
Quick
10 questions, about 20 minutes
Flashcards
Exam tips
What people who passed say about CCAR-F. These are candidate reports, not official guidance, so check the official exam guide too.
- It's built around production scenarios. Each sitting draws four of six, such as a support agent that hands off to people, a multi-agent research system, Claude Code in a CI pipeline, and pulling structured data out of documents. Read the scenario before its questions.
- Pace yourself. You get about two minutes per question and the scenarios are long. Answer the ones you're sure of, flag the rest and come back.
- Common traps:
- retrying errors that can never succeed on a retry
- letting an agent guess when it should escalate to a person
- treating "reply in JSON" in a prompt as if it were schema-enforced output
- trusting a model's own confidence score instead of checking in code
- assuming your personal settings override project or company settings
- taking human review out of high-stakes actions
- Build, don't just read. Wire up a coordinator with subagents, a small MCP server, and Claude Code hooks and settings. Scenario questions reward people who have hit these problems.
Sources: Very Good Ventures, Level Up Coding, CertLand
Section drills
Practice every question in one domain, in review mode. Weights show how much each domain counts on the real exam.
| Domain | Weight | Questions | Your accuracy | Drill |
|---|---|---|---|---|
| D1 Agentic Architecture & Orchestration | 27% | 98 | Not tried | Drill |
| D2 Claude Code Configuration & Workflows | 20% | 73 | Not tried | Drill |
| D3 Prompt Engineering & Structured Output | 20% | 73 | Not tried | Drill |
| D4 Tool Design & MCP Integration | 18% | 67 | Not tried | Drill |
| D5 Context Management & Reliability | 15% | 55 | Not tried | Drill |
About this exam
Claude Certified Architect – Foundations was Anthropic's first technical certification, launched in March 2026. Each question gives you a project scenario and asks for the best architectural decision, so it rewards hands-on experience over memorized facts.
The exam is proctored and closed-book, and Pearson VUE delivers it. Registration has been limited to people at Claude Partner Network organizations, so check your eligibility before you plan a date.
The heaviest domain is agentic architecture and orchestration. Expect scenarios about when to split work across subagents, how to enforce rules with hooks rather than instructions, how to design tools and MCP servers, and how to keep long-running agents reliable.