Work at a Claude Partner Network organization
RequiredRegister with your organization's recognized company email domain. You must be 18 or older.
Claude Certified Architect – Foundations (CCAR-F) Exam Guide
Validates knowledge of designing production Claude solutions, including agent orchestration, tool integration, Claude Code, prompting and context management.
The exam tests design judgment across real production scenarios: which agent pattern, tool design, Claude Code setting or prompting approach is better, and why. Use the exam guide as your syllabus, take the free technical prep courses, build the guide's scenarios yourself and finish with exam-level practice questions.
Understand the exam
Syllabus
1Agentic architecture & orchestration27%
- 1.1Design and implement agentic loops for autonomous task execution
- 1.2Orchestrate multi-agent systems with coordinator-subagent patterns
- 1.3Configure subagent invocation, context passing, and spawning
- 1.4Implement multi-step workflows with enforcement and handoff patterns
- 1.5Apply Agent SDK hooks for tool call interception and data normalization
- 1.6Design task decomposition strategies for complex workflows
- 1.7Manage session state, resumption, and forking
Example scenario: A support agent sometimes issues refunds before looking up the customer. Would you fix this with a prompt instruction or a programmatic gate, and why?
2Tool design & MCP integration18%
- 2.1Design effective tool interfaces with clear descriptions and boundaries
- 2.2Implement structured error responses for MCP tools
- 2.3Distribute tools appropriately across agents and configure tool choice
- 2.4Integrate MCP servers into Claude Code and agent workflows
- 2.5Select and apply built-in tools (Read, Write, Edit, Bash, Grep, Glob) effectively
3Claude Code configuration & workflows20%
- 3.1Configure CLAUDE.md files with appropriate hierarchy, scoping, and modular organization
- 3.2Create and configure custom slash commands and skills
- 3.3Apply path-specific rules for conditional convention loading
- 3.4Determine when to use plan mode vs direct execution
- 3.5Apply iterative refinement techniques for progressive improvement
- 3.6Integrate Claude Code into CI/CD pipelines
Example scenario: Frontend and backend folders in one repository need different coding rules. Where would you put each set of rules?
4Prompt engineering & structured output20%
- 4.1Design prompts with explicit criteria to improve precision and reduce false positives
- 4.2Apply few-shot prompting to improve output consistency and quality
- 4.3Enforce structured output using tool use and JSON schemas
- 4.4Implement validation, retry, and feedback loops for extraction quality
- 4.5Design efficient batch processing strategies
- 4.6Design multi-instance and multi-pass review architectures
5Context management & reliability15%
- 5.1Manage conversation context to preserve critical information across long interactions
- 5.2Design effective escalation and ambiguity resolution patterns
- 5.3Implement error propagation strategies across multi-agent systems
- 5.4Manage context effectively in large codebase exploration
- 5.5Design human review workflows and confidence calibration
- 5.6Preserve information provenance and handle uncertainty in multi-source synthesis
Example scenario: A research agent's final report drops which source each claim came from. How would you keep provenance through the pipeline?
Official exam guide · Version 1.0, July 2026
Format
- Questions
- 60
- Duration
- 2h
- Passing score
- 720 / 1000
- Delivery
- Test centre or online
- Languages
- English
Question types
Choose the single best answer; each question states how many answers to select.
CertUp example
A coordinator agent hands research to a synthesis subagent, but the synthesis subagent's reports ignore earlier search results. What is the most likely cause?
Show answer
C. Subagents start with isolated context, so the coordinator must include prior findings in the subagent's prompt.
Select the stated number of correct answers from the list.
CertUp example
Select TWO changes that make an extraction pipeline more reliable when Claude sometimes returns malformed output.
Show answer
A and B. A schema constrains the format, and a validation-retry loop feeds the specific error back for correction; a vague instruction or a higher cap does neither.
Eligibility and booking
Sign in to the Anthropic Partner Academy with your partner company email, register and pay, then schedule with Pearson VUE. Your registration stays valid for 5 years.
Make a study plan
Read the exam guide first, including its scenarios and sample questions, and mark the domains you use least.
Most candidates spend 15–40 hours. Plan for the low end if you use Claude Code and the Agent SDK daily; plan for the high end if you have used Claude for about six months but not all of the exam domains.
- 1Understand the examRead the syllabus and format, check you’re eligible and pick a target date.2 h
- 3Learn the core conceptsWork through one main course or book, topic by topic.17 h
- 4Practice with mock examsTake timed practice tests and review every mistake.9 h
- 5Prepare for exam dayBook the exam and check the ID and delivery rules.2 h
Learn the core concepts
Anthropic lists free prep courses on the Partner Academy. The three technical courses map most directly to the domains.
Best official startBuilding with the Claude APIBest for MCPIntroduction to Model Context ProtocolBest for Claude CodeClaude Code in ActionBuild the scenarios The exam asks which design is better, not which works. Building each guide scenario yourself shows you the tradeoffs.
Practice with mock exams
Start with the official sample questions, then use exam-level question sets and review why each wrong option loses.
CCAR-F Exam Guide sample questions
CCAR-F Practice Questions
Claude Certified Architect Foundations mock examAre you ready?
- You can explain why each wrong option in a practice question is worse, not just which one is right
- You know when to enforce a rule with a hook or gate rather than a prompt
- You can choose between CLAUDE.md, rules files, skills and hooks for a given need
Prepare for exam day
Use the checklist for your delivery route.
Before testing from home
Before travelling to the test centre
- Run the OnVUE system test on the computer and network you will use
- Clear your desk and keep phones, extra monitors, notes and AI tools out of reach
- Stay in webcam view for the whole session
- Arrive early to check in
- Bring a valid government photo ID whose name exactly matches your registration
- Plan for about 135 minutes in total, including check-in
- Reschedule or cancel in your Pearson VUE account at least 48 hours before your appointment. Later changes or a missed appointment forfeit the exam fee
- If you fail, wait 14 days before the second attempt, 30 before the third and 90 before the fourth, with up to 4 attempts in 12 months
Advice from exam takers
It is a judgment exam
A September 2026 passer (833/1000) found that most questions offer four plausible options and the task is picking the more efficient, deterministic or maintainable one. Reading the stem closely before the options mattered most.
Daily Claude Code use counts for a lot
The same passer studied about 16 hours after a year of daily Claude Code use, and found exam-level practice tests the most useful preparation. Another candidate passed after two weeks of focused study.
Ask where a failure would land
An independent breakdown of a common question type suggests a test: if downstream code depends on the result or a mistake reaches production, choose a mechanism such as a hook or gate; if the failure stays in the conversation, prompting is enough.