The Claude Certified Architect Professional (CCAR-P) is Anthropic's advanced credential for architects who design, build and deliver production AI systems on Claude. It is a proctored exam with 63 questions in 120 minutes and costs $175. You need a scaled 720/1,000 to pass. It covers seven domains, from RAG and integration to governance and stakeholder management.
I checked every fact in this guide against Anthropic's official CCAR-P Exam Guide (version 1.0, effective July 2026), the Anthropic Partner Academy certifications page and the Pearson VUE Anthropic page. Most other pages about this exam are short and skip the parts that matter. This one covers what each domain really asks of you, how the exam differs from the Claude Certified Architect Foundations exam, and a study plan you can follow.
CCAR-P exam: key facts at a glance
Exam code | CCAR-P (sometimes written CCA-P) |
|---|---|
Issued by | Anthropic (Claude Certification Program) |
Number of questions | 63 multiple-choice and multiple-response items. Each item tells you how many answers to select. |
Structure | Standalone items, not grouped into scenarios |
Time limit | 120 minutes |
Passing score | Scaled 720 on a 100–1,000 scale |
Cost | $175 USD per attempt. Partner-tier discounts may apply at checkout. |
Delivery | Pearson VUE: online proctored (OnVUE) or at a test center |
Validity | 12 months. On-time renewal is free. |
Prerequisites | None required. You do not need Foundations first. |
Recommended experience | 3+ years in systems architecture or platform engineering, plus 6+ months with Claude or comparable LLMs in production |
Eligibility | Organizations in the Claude Partner Network (membership is free) |
What is the Claude Certified Architect Professional?
The Claude Certified Architect – Professional is the senior architect credential in Anthropic's Claude Certification Program. According to the official guide, it shows that the holder can own, or contribute heavily to, the full lifecycle of a Claude-powered system, from the first discovery workshop through to monitoring in production.
It goes well beyond the technical detail of the Foundations exam. Anthropic lists seven things a certified Professional can do:
Design and prototype AI solutions that solve real business problems
Choose the right Claude models, architectures and API patterns for each use case
Apply prompt engineering and context strategies to steer model behavior
Integrate Claude into production systems through APIs, orchestration tools and data pipelines
Use evaluation, monitoring and observability to keep quality high
Build security, compliance and governance into the design
Work with cross-functional stakeholders and explain architectural decisions clearly
In short, Foundations asks whether you can build it correctly. Professional asks whether you can design it, defend it, govern it and keep it running in a real company.
Who the Claude certified professional exam is for (and who it isn't for)
The guide is unusually direct about who should take this exam.
It is for mid-level to senior technical professionals, mainly solution architects, AI/ML engineers, technical leads and senior software engineers who sit between business requirements and technical implementation. They often advise clients or internal teams and lead architecture discussions that involve security, legal and executive stakeholders. The guide mentions industries such as financial services, healthcare, retail, technology, education and government.
It is not for entry-level developers, casual users of Claude apps, people without experience designing end-to-end AI systems, or roles limited to writing prompts with no wider design responsibility.
Recommended experience (not mandatory):
A foundation in software engineering best practices: modular design, separation of concerns and scalability
3+ years in systems architecture or platform engineering
6+ months of hands-on work with Claude or comparable LLM systems in production
Experience taking systems from discovery through deployment and into operation
There are no mandatory prerequisites. You do not need to hold the Claude Certified Architect – Foundations credential first. The credential is awarded on exam performance alone.
Who can take the CCAR-P?
Like every exam in the program, the Professional exam is currently open to people at organizations in the Claude Partner Network. Membership is free and open to any organization bringing Claude to market, such as consultancies, agencies, software vendors and systems integrators. If your company is not a partner yet, ask it to apply. You can study the free Anthropic Academy courses in the meantime.
Claude Certified Architect Professional exam format and scoring
63 items in 120 minutes, which is just under 2 minutes per question.
Multiple-choice and multiple-response. Each item tells you how many answers to pick.
Standalone items, not scenarios. This is the biggest format difference from Foundations, which groups its questions under 4 of 6 published scenarios. On the Professional exam every question sets its own context, so you cannot prepare for a fixed set of case studies. You need to cover the whole blueprint.
Criterion-referenced scoring. You are measured against a fixed standard that experts set for a minimally qualified candidate, not against other candidates. The cut score is a scaled 720. A scaled 720 does not mean you need 72% of answers correct.
Score report. You get pass or fail, a scaled score from 100 to 1,000, and the percentage you got right in each domain. Only the total scaled score decides pass or fail.
What the questions reward: the official sample items all have the same shape. Several answers sound reasonable, and the best one fixes the root cause with the smallest, most principled change. That means removing unneeded capabilities rather than monitoring them, ordering a prompt so it can be cached rather than truncating it, and checking retrieval before blaming the model.
The 7 CCAR-P exam domains (with weights)
# | Domain | Weight | ≈ Questions |
|---|---|---|---|
1 | Solution Design & Architecture | 17% | ~11 |
2 | Claude Models, Prompting & Context Engineering | 13% | ~8 |
3 | Integration | 19% | ~12 |
4 | Evaluation, Testing & Optimization | 16% | ~10 |
5 | Governance, Safety & Risk Management | 14% | ~9 |
6 | Stakeholder Communication & Lifecycle Management | 14% | ~9 |
7 | Developer Productivity & Operational Enablement | 7% | ~4 |
Domain 1: Solution Design & Architecture (17%)
Turn business problems into Claude solutions. Start from the business outcome, not the model.
Design end-to-end architectures that run input → processing → output → feedback loop. Questions often ask what is missing from a design, and the answer is frequently the feedback loop.
Choose the pattern: a workflow (fixed, predictable steps), an agent (the model chooses its next steps), or an augmented LLM (one call plus retrieval and tools). Use the simplest pattern that solves the problem.
Multi-agent systems and orchestration, and breaking complex problems into smaller parts.
Tie the design to business value: efficiency, transformation, productivity, cost and performance SLAs.
Domain 2: Claude Models, Prompting & Context Engineering (13%)
Model selection by trade-off: capability against latency against cost. Pick a model that fits the task, not automatically the largest or the cheapest.
System prompts, templates and guardrails.
Prompting techniques: zero-shot, few-shot and chain-of-thought, and when each one is worth the extra tokens.
Context window and token management.
Prompt reuse: prompt caching (put stable content first), modular prompts and Skills.
Domain 3: Integration (19%)
This is the heaviest domain and has the most tasks, eight in all.
Capability bloat: spot agents with more tools or permissions than their role needs, and apply least privilege.
Authentication and authorization gaps: who and what the agent can act as.
Accuracy-latency trade-offs: justify configuration choices with numbers.
Observability at scale: choose monitoring strategies for systems with many calls and many agents.
RAG pipeline design: chunking and indexing strategies.
Retrieval matched to the data and the queries: for example, keyword search for exact identifiers, semantic search for questions written in natural language, and hybrid search when users mix the two.
Choosing the integration mechanism: MCP, API/CLI or agent-to-agent.
Progressive discovery versus monolithic context: load information when it is needed instead of putting everything in the prompt up front.
Domain 4: Evaluation, Testing & Optimization (16%)
Metrics for accuracy, latency, cost, safety and security.
Evaluation datasets and test frameworks that mix methods: automated checks, model-graded scoring and human review.
A/B testing and iterative improvement.
Diagnosis: tell a prompt failure apart from a hallucination and from a model that is wrong for the task.
Optimizing token use, latency and cost against performance.
Monitoring with logging and observability tools.
Domain 5: Governance, Safety & Risk Management (14%)
Guardrails and safety controls
The risks, limitations and failure modes of LLM systems
Human-in-the-loop validation
Regulatory compliance: the guide names GDPR, HIPAA and FedRAMP as examples
Ethical AI: bias, fairness and transparency
Domain 6: Stakeholder Communication & Lifecycle Management (14%)
Structured discovery and requirements gathering
Explaining architectural decisions and trade-offs to technical and non-technical audiences
Managing feedback loops and expectations, including SLAs
Documenting the architecture and giving implementation guidance
Supporting every lifecycle phase: discovery, design, handoff, monitoring and iteration
Domain 7: Developer Productivity & Operational Enablement (7%)
Configuring Claude tools and environments for teams, for example Claude Code
Improving developer workflows with AI-assisted tools
Supporting debugging and the resolution of operational issues
Claude Certified Architect Foundations vs Professional
Foundations (CCAR-F) | Professional (CCAR-P) | |
|---|---|---|
Questions | 60 | 63 |
Structure | 4 scenarios drawn from a bank of 6 | Standalone items |
Time | 120 minutes | 120 minutes |
Passing score | 720 / 1,000 | 720 / 1,000 |
Fee | $125 | $175 |
Domains | 5 | 7 |
Focus | Building it correctly: Agent SDK loops, Claude Code configuration, MCP tool design, structured output | Designing and running it end to end: RAG, integration, evaluation, governance and compliance, stakeholders |
Recommended experience | 6+ months building with Claude | 3+ years in architecture, plus 6+ months with LLMs in production |
Prerequisites | None | None (Foundations is not required) |
Validity | 12 months, free renewal | 12 months, free renewal |
Which should you take first?
If you build day to day, writing agents, MCP servers and Claude Code setups, start with Foundations. It is cheaper, covers the concrete material, and gives you most of the technical depth that Professional assumes.
If you lead client engagements or own architecture decisions and already have the recommended experience, you can go straight to Professional.
Doing both makes sense. The two exams overlap only a little. Foundations is about mechanics, and Professional is about judgment across the whole system and the organization around it.
CCAR-P exam cost, retakes and policies
The Claude Certified Architect Professional exam costs $175 USD per attempt. The price shown at checkout reflects any discount your partner tier qualifies for. For comparison, the Partner Academy lists the Associate – Foundations exam at $99, and the Developer – Foundations and Architect – Foundations exams at $125 each.
Retakes: you must wait 14 days after a first fail, 30 days after a second and 90 days after a third. The limit is 4 attempts in a rolling 12 months, and each attempt costs the full fee. The limits apply per exam, so failing Professional does not stop you booking Foundations.
Rescheduling: the exam guide says you can cancel or reschedule up to 24 hours before your appointment. Pearson VUE's Anthropic page says 48 hours. Plan on 48 hours to be safe. A late change or a no-show means you lose the fee.
ID: the name on your registration must exactly match your government photo ID. To correct your name, email certifications-support@anthropic.com before you schedule.
Exam rules: no notes, phones, smartwatches or second monitors. You also accept a non-disclosure agreement, so avoid anyone selling "real exam questions."
Renewal: the credential lasts 12 months. On-time renewal is a free, non-proctored assessment. If your credential lapses, you retake the full exam at the full fee. If the exam content changes significantly, Anthropic may require a full retake.
Appeals: submit an appeal within 14 days through Pearson VUE.
How to become a Claude Certified Architect Professional
Check your eligibility and experience. Your organization must be in the Claude Partner Network. Then be honest about the recommended 3+ years in architecture and 6+ months running LLM systems in production.
Assess yourself against the blueprint. Download the CCAR-P exam guide from the Anthropic Partner Academy and rate yourself from 1 to 5 on every task in the seven domains.
Build one end-to-end Claude solution. Anthropic specifically recommends a system that includes RAG, evaluation and observability.
Practice architectural decisions: model selection, integration protocols and security trade-offs. Practice explaining each choice to a non-technical stakeholder.
Register and schedule. Buy the exam ($175), create your Pearson VUE account and book an online or test-center slot.
Pass and renew. You need 720 or more. Share your Credly badge, and set a reminder to renew for free before 12 months are up.
CCAR-P study guide: a 6-week plan
This plan assumes you already meet the recommended experience. It is ordered by domain weight, so the domains that carry the most marks come first.
Week | Domains | Hands-on deliverable |
|---|---|---|
1 | 3: Integration (19%) | A RAG pipeline over a real document set. Compare two chunking strategies and keyword, semantic and hybrid retrieval. Expose one internal system through MCP. |
2 | 1: Solution Design (17%) | Write a one-page architecture for a business case. Justify choosing a workflow, an agent or an augmented LLM, and include a feedback loop and the business value metrics. |
3 | 4: Evaluation (16%) | Build an evaluation set covering typical cases, edge cases and adversarial cases. Run an A/B comparison of two prompts or models, and track accuracy, latency and cost. |
4 | 5: Governance (14%) and 6: Stakeholder (14%) | Add guardrails, a human-in-the-loop review step and a data-handling note for GDPR or HIPAA. Write a decision record and an SLA proposal for a stakeholder. |
5 | 2: Models & Prompting (13%) and 7: Developer Productivity (7%) | Add prompt caching to your system and measure the difference. Compare two model tiers. Set up a shared Claude Code configuration for a team. |
6 | Review | Redo the self-assessment and work through the official sample questions. Re-study every task you rated 3 or lower. |
Free resources
Anthropic Academy: free courses on the Claude API, MCP, Claude Code and agents
Claude documentation: models, prompt engineering, prompt caching, tool use and Skills
Model Context Protocol docs: integration patterns and security
Our Claude Certified Architect Foundations guide, for the hands-on material that Professional assumes
Claude Certified Architect Professional practice questions
These five questions are original practice items written for this guide. They are not real exam questions and are not the official samples. They follow the style of the official guide: several plausible options, and one best answer.
Question 1: Integration
A support assistant answers questions from 40,000 product manuals. Users often ask about exact part numbers such as "XR-4471-B", and pure semantic search keeps returning manuals for similar but wrong parts. What is the best fix?
Switch to a larger embedding model
Use hybrid retrieval that combines keyword (lexical) matching with semantic search, so exact identifiers are matched precisely
Put all 40,000 manuals into the context window
Tell users to describe the part in words instead of using its number
Show answer
B. Retrieval should match both the data and the kind of query. Exact identifiers need lexical matching, and natural-language questions need semantic search. A bigger embedding model does not fix the mismatch, option C is neither feasible nor cost-effective, and option D pushes the problem onto users.
Question 2: Solution Design
A finance team wants to sort incoming invoices into 12 fixed categories and send each one to a known approval queue. The steps never change. Which pattern fits best?
An autonomous multi-agent system that decides its own steps
A workflow: a classification call with a structured output, followed by deterministic routing code
A long-running agent with access to every finance tool
A fine-tuned custom model built before any prototype
Show answer
B. When the steps are predictable, a workflow is cheaper, faster, easier to test and easier to audit than an agent. Agents are for open-ended tasks where the path is not known in advance.
Question 3: Evaluation
The team wants to swap to a cheaper model to cut costs, and the product owner asks whether quality will drop. What is the best way to decide?
Have three engineers read ten outputs from each model
Run both models on the same representative evaluation set, including edge cases, and compare accuracy, latency and cost. Then confirm the result with a limited A/B test in production.
Switch over and wait for user complaints
Pick the model with the higher score on a public leaderboard
Show answer
B. Decisions should rest on your own task-specific evaluation data across several metrics, followed by a controlled rollout. Small samples, waiting for complaints and generic leaderboards do not measure your workload.
Question 4: Governance
A hospital wants Claude to draft summaries of clinical notes that will be stored in patient records. The design must be HIPAA-aware and safe. Which control matters most?
Add a disclaimer at the bottom of every summary
Set the temperature to 0
Require a clinician to review and approve each summary before it is written to the record, and keep PHI handling within approved, compliant infrastructure
Use the largest model available
Show answer
C. In high-risk, regulated settings, human-in-the-loop validation and compliant data handling are the main controls. A disclaimer, temperature and model size do not stop a wrong summary from entering a medical record.
Question 5: Stakeholder Communication
Before launch, an executive asks you to guarantee that the AI assistant will be "100% accurate." What is the best response?
Agree, to keep the project on schedule
Explain that LLM outputs are probabilistic, then agree measurable targets on an evaluation set, an SLA, human review for low-confidence cases, and a phased rollout
Refuse to launch until accuracy reaches 100%
Leave the topic out of the project documentation
Show answer
B. Architects manage expectations openly and turn vague demands into measurable, agreed targets with mitigations in place. The other options either over-promise or stop delivery.
Is the Claude Certified Architect Professional worth it?
Yes, if you are the person clients or leadership trust with Claude architecture decisions. Of all the Claude professional certifications available today, it is the one that best signals senior, end-to-end ownership.
It proves seniority. Foundations shows you can build. Professional shows you can design, govern and deliver, which is what consultancies charge senior rates for.
Few people hold it. The official exam guide was first published in July 2026, and search interest is just starting to grow.
It counts toward Partner Network standing for your organization.
The blueprint covers real enterprise work. RAG, evaluation, compliance and SLAs are what decide whether an AI project ships or stalls.
The cost is low. $175, with free renewal.
Downsides: it is limited to Partner Network organizations, it has to be renewed every year, there is no published scenario bank to study from, and employers do not yet recognize it as widely as established cloud architect certifications.
Claude Certified Architect Professional FAQ
How much does the CCAR-P exam cost?
$175 USD per attempt. Partner-tier discounts may apply, and on-time renewal is free.
Do I need Foundations before the Professional?
No. There are no mandatory prerequisites. Anthropic recommends 3+ years in architecture and 6+ months with Claude or comparable LLMs in production.
How many questions are on the exam?
63 multiple-choice and multiple-response items in 120 minutes. Each one is standalone rather than part of a scenario.
What is the passing score?
A scaled 720 on a 100–1,000 scale.
What is the difference between Foundations and Professional?
Foundations ($125, 60 items, 5 domains, scenario-based) tests hands-on building with Claude Code, the Agent SDK, the API and MCP. Professional ($175, 63 items, 7 domains) tests end-to-end enterprise architecture, evaluation, governance and stakeholder management. Read our Claude Certified Architect Foundations guide for the full details on Foundations.
Who can take it?
People at organizations in the Claude Partner Network. Membership is free.
How long is it valid?
12 months. Renew for free with a non-proctored assessment.
Which domain has the highest weight?
Integration at 19%, followed by Solution Design & Architecture at 17% and Evaluation, Testing & Optimization at 16%.
What happens if I fail?
You can retake the exam after 14, 30 and then 90 days, with up to 4 attempts in a rolling 12 months. Each attempt costs the full fee.
Is CCA-P the same as CCAR-P?
Yes. The official exam code is CCAR-P. Some sites shorten it to CCA-P.
Final word
The Claude Certified Architect Professional exam tests the judgment of someone who has taken a Claude system from a whiteboard to production and kept it running. Integration, evaluation and governance carry nearly half the marks. Build one complete system with RAG, evaluations and guardrails, practice defending each decision, and you will be ready for the exam.