Claude Certified Architect Certification: Complete Guide (2027)
What the Claude Certified Architect certification is, who can take it, how the CCA-F exam works, what it covers, whether it's worth it, and how to prepare, checked against Anthropic's official exam guide and FAQ.
The Claude Certified Architect – Foundations certification is Anthropic's credential for solution architects who design and build agent systems with Claude. You earn it by passing a proctored, scenario-based exam (exam code CCAR-F, often shortened to CCA-F) that tests the production decisions behind reliable Claude systems, across the Claude API, the Claude Agent SDK, Claude Code, and Model Context Protocol (MCP).
This guide covers what the certification is, who can take it, how the exam works, what it covers, whether it's worth your time, and how to prepare. The exam details come from Anthropic's official exam guide (version 1.0, July 2026), the certification pages and FAQ on the Anthropic Partner Academy, and Anthropic's July 2026 program announcement.
The short version, if you only need the facts:
The exam has 60 multiple-choice and multiple-response questions, and you get 120 minutes.
You pass with a scaled score of 720 on a range of 100 to 1,000.
It costs $125 USD, before any partner-tier discount.
Only people at Claude Partner Network organizations can register.
The credential lasts 12 months, and renewing on time is free.
What is the Claude Certified Architect certification?#
It's one of four credentials in the Claude Certification Program, which Anthropic launched in March 2026 alongside the Claude Partner Network, its program for the consulting and services firms that deploy Claude for customers. Anthropic backed the network with a $100 million investment and set out to make its certifications a benchmark for verified expertise. Every exam is proctored and identity-verified, and Anthropic says the exams were designed by subject matter experts, including its own Applied AI team.
Adoption was fast. By July 2026, more than 36,000 consultants at more than 1,300 organizations had earned a Claude certification, according to Anthropic. The program has also changed several times since launch, so check dates when you read older prep material.
Two of those changes matter most if you're studying now. Since June 30, 2026, exams are delivered by Pearson and badges are issued through Credly, and a certification stays valid for 12 months instead of six. And on December 7, 2026, the Associate, Developer, and Architect – Professional exams switch to updated content, with updated exam guides available from November 17. Architect – Foundations isn't part of that update, so its July 2026 guide is still the one to study.
The four Claude certifications and where Architect fits#
Certification | Who it's for | Price (USD) | Counts toward partner tier |
Associate – Foundations | Everyday, practical use of Claude, for consultants, project leads, and other business or technical roles | $99 | No |
Developer – Foundations | Engineers building applications with the Claude API, tool use, and agents | $125 | Yes |
Architect – Foundations | Solution architects who design and build agent systems with Claude | $125 | Yes |
Architect – Professional | Integration architecture, governance, and evaluation at large-enterprise scale | $175 | Yes |
The Developer and Architect – Foundations exams cost the same and both cover agents, so the domain weights are the clearest way to choose. Developer puts the most weight on applications and integration (33.1%) and on model selection and optimization (16.8%). Architect – Foundations puts the most weight on agentic architecture and orchestration (27%), then Claude Code configuration and workflows, and prompt engineering and structured output (20% each). If your job is deciding how an agent system should be structured and how it stays reliable, Architect – Foundations is the better match.
Foundations and Professional are separate certifications with separate exams. There's no formal prerequisite, so you can take Professional without Foundations, but Anthropic calls Foundations the natural place to start, and passing it doesn't upgrade you to Professional.
Who can take the exam#
This is the detail most guides skip. According to the Partner Academy FAQ, certification is currently available only to people at organizations in the Claude Partner Network. You register with your company email, and the domain has to be recognized on your firm's partner record. Personal email addresses don't work, and customers or other non-partners can't register for now.
A few practical consequences follow.
If you work at a consultancy, systems integrator, or services firm that builds on Claude, ask whoever manages the partnership whether your firm has joined. Firms can join the network for free at claude.com/partners.
If your email domain isn't recognized, Anthropic's partner support team has to add it to your firm's record, which takes 7 to 10 days. Ask early if your company uses several domains.
You need to be at least 18, and you'll show a government-issued ID at check-in.
If you aren't eligible yet, you can still prepare. The exam guide is a public download, Anthropic publishes free courses through its Academy, and the skills are the same ones production Claude work demands.
The CCA-F exam at a glance#
Detail | What Anthropic says |
Exam code | CCAR-F |
Questions | 60 multiple-choice and multiple-response items. Each item states how many answers to select. |
Structure | 4 scenarios, drawn at random from a bank of 6 |
Time | 120 minutes, about 135 minutes of total seat time with check-in |
Passing score | 720 on a scaled range of 100 to 1,000 |
Delivery | Online proctored through Pearson or at a Pearson test center |
Languages | English or Japanese, chosen when you schedule |
Fee | $125 USD before any partner-tier discount |
Results | On screen when you finish, followed by a Credly badge if you pass |
Validity | 12 months, with a free renewal assessment if you renew on time |
Because the score is scaled, 720 doesn't necessarily mean 72% of questions answered correctly. Scaling lets Anthropic compare results across exam forms that differ slightly in difficulty. Your score report also shows the percentage you got right in each domain, which tells you what to review if you need a retake.
What the exam covers#
The blueprint has five domains. Each weight is the approximate share of scored questions drawn from that domain.
Agentic Architecture & Orchestration (27%) covers the agent loop itself, coordinator and subagent designs, passing context to subagents, enforcing workflow rules with hooks, breaking complex tasks down, and deciding when to resume, fork, or restart a session.
Claude Code Configuration & Workflows (20%) covers the
CLAUDE.mdhierarchy, custom slash commands and skills, path-specific rules, plan mode versus direct execution, iterative refinement, and running Claude Code in CI/CD pipelines.Prompt Engineering & Structured Output (20%) covers explicit review criteria, few-shot examples, structured output through tool use and JSON schemas, validation and retry loops, batch processing, and multi-pass review designs.
Tool Design & MCP Integration (18%) covers tool descriptions that help Claude choose between similar tools, structured error responses, how tools are distributed across agents, MCP server configuration, and the built-in tools such as Read, Edit, Bash, and Grep.
Context Management & Reliability (15%) covers keeping critical facts intact across long sessions, escalation and ambiguity handling, error propagation between agents, large-codebase exploration, human review workflows, and provenance when you combine sources.
The guide expects hands-on familiarity. Its ideal candidate is a solution architect with six or more months of practical experience building with the Claude API, the Agent SDK, Claude Code, and MCP.
Six scenarios, four per sitting#
Every question sits inside a scenario that describes a realistic production system, and your sitting draws four of the six at random.
Customer support resolution agent, an Agent SDK agent that handles returns, billing disputes, and account issues through custom MCP tools and has to know when to escalate.
Code generation with Claude Code, where a team uses Claude Code for generation, refactoring, debugging, and documentation, with custom commands and
CLAUDE.mdconfiguration.Multi-agent research system, in which a coordinator delegates to subagents that search, analyze, synthesize, and write cited reports.
Developer productivity with Claude, an agent that helps engineers explore unfamiliar codebases using the built-in tools and MCP servers.
Claude Code for continuous integration, where Claude Code runs automated reviews, generates tests, and comments on pull requests.
Structured data extraction, a system that pulls data from unstructured documents and validates it against JSON schemas.
The matrix shows why skipping a domain is risky. Context Management & Reliability carries the smallest weight (15%), yet it's a primary domain in four of the six scenarios. Prompt Engineering & Structured Output carries 20% but is listed as primary for only two, the CI pipeline and structured extraction.
What kind of thinking the exam rewards#
The guide says the exam tests "practical judgment about architecture, configuration, and tradeoffs in production deployments," and its task statements and sample questions keep returning to three habits.
Guarantees belong in code. When a business rule has to hold every time, such as verifying a customer before a refund, the guide favors programmatic enforcement like hooks and prerequisite checks, because prompt instructions alone have a non-zero failure rate.
Machine signals beat prose. An agent loop should continue while
stop_reasonis"tool_use"and end on"end_turn". The guide lists parsing Claude's wording, arbitrary iteration caps, and checking for assistant text as anti-patterns.The first fix should be proportionate. The sample questions reward the change that addresses the root cause with the least machinery, such as rewriting vague tool descriptions before building a routing layer. Over-engineered options are usually the wrong ones.
Candidates accept a non-disclosure agreement, so real exam questions aren't something anyone can share. Here is a practice question we wrote in the style of the guide's samples. It is not an exam item.
Your invoice extraction pipeline uses tool use with a strict JSON schema, and every output parses. A spot check finds the wrong vendor name in 6% of invoices. What should you add first?
A. A stricter JSON schema with an enum of known vendors
B. Validation logic that checks extracted values against the source document and retries with specific feedback
C. A higher max_tokens limit so the model has room to reason
D. A larger context window
The answer is B. The schema already guarantees structure, and these errors are about meaning, so they need a separate check plus a retry that tells Claude exactly what was wrong. A is tempting, but an enum would break on every vendor you haven't seen yet. C and D don't address the failure.
Is the Claude Certified Architect certification worth it?#
For people at partner firms, the case is strong. Certified practitioner counts determine a firm's standing in the Claude Partner Network, and the top tier, Global Premier, requires 1,000 certified practitioners. Anthropic says firms are prioritizing certification in hiring, and the largest ones have made public commitments.
A few things to weigh before you commit.
Access. You can only sit the exam through a partner firm, so it's most useful to consultants, solution architects, and forward-deployed engineers at those firms.
Upkeep. The credential lasts 12 months. On-time renewal is a free assessment, but a lapse means retaking the full exam at the full fee.
Scope. It certifies judgment about Claude specifically. If your work spans several model providers, treat it as one part of a broader AI architecture profile.
Cost. The fee is $125 before discounts, and your firm's partner tier may lower it.
How to prepare#
Read the official exam guide first. Anthropic calls it the authoritative source for exam scope. It lists every task statement, the six scenarios, sample questions, and preparation exercises, and it's available from the certification page.
Build what the guide tells you to practice. Its preparation exercises are the best checklist. Build an agent with the Agent SDK that includes tool calling, error handling, and subagents. Configure Claude Code for a real project with a
CLAUDE.mdhierarchy, path-specific rules, custom skills, and an MCP server. Design MCP tools with clearly different descriptions and structured errors. Build an extraction pipeline with JSON schemas and a validation-retry loop.Take Anthropic's free courses. The certification page lists free Anthropic Academy courses that cover the exam domains, including Claude 101, Building with the Claude API, Introduction to Model Context Protocol, and Claude Code in Action. Anthropic's public Academy is at academy.claude.com.
Practice scenario questions under time. Anthropic retired its practice exam when certification moved to Pearson, so the guide's sample questions are the only official practice. Sixty questions in 120 minutes leaves about two minutes per question, scenario reading included.
Plan around the score report. If you need a retake, the per-domain percentages tell you where to spend the waiting period.
The plan above assumes you already work with Claude and can give the exam about five hours a week.
Week 1. Start with agentic architecture, the heaviest domain. Read the guide end to end, then build an agent loop with one tool, a guardrail in code, and structured error responses.
Week 2. Move on to tools, MCP, and context. Add a coordinator with two subagents, pass context to them explicitly, and connect one MCP server whose tools have distinct descriptions.
Week 3. Cover Claude Code and prompting. Configure Claude Code with a
CLAUDE.mdhierarchy, a path-specific rule, and a custom skill, run it non-interactively in CI, and build an extraction pipeline with a validation-retry loop.Week 4. Practice the scenarios against the clock. Work through the guide's sample questions and timed scenario sets, then give the rest of the week to your weakest domain.
How to register and what to expect on exam day#
Confirm you're eligible. Your company email has to be on a Claude Partner Network firm's record, and you have to be in a country Anthropic supports.
Sign in to the Anthropic Partner Academy with that email, open the CCA-F certification page, download the exam guide, and read the certification terms and exam policy.
Select Register and confirm your name. It has to match your government-issued ID exactly, or Pearson won't let you test, and you can change it yourself only once a year.
Finish at Pearson, where you arrive already signed in. Choose online proctoring or a test center, pick a language, date, and time, and pay at checkout. A partner discount, if your firm has one, is applied automatically.
Since November 12, 2026, you register, schedule, and pay in one pass, with no separate Pearson account to set up. The scheduling rules are strict. You can reschedule or cancel up to 48 hours before a test center appointment or up to one hour before an online exam, and a no-show forfeits the fee. If you paid at Pearson's checkout, cancelling starts an automatic refund. If you need accommodations, request them through Pearson before you schedule, at least 10 days ahead. If English is your second language, you can ask Pearson for extra time. Browser translation tools aren't allowed during the exam, so pick the English or Japanese version you're most comfortable with.
You'll see your score on screen when you finish, and if you pass, Credly emails you to claim your badge. If you don't pass, you can retake after 14 days. The wait grows to 30 days after a second failed attempt and 90 days after a third, and you can sit the exam up to four times in a rolling 12-month period. Each retake costs the full fee, although your partner-tier discount still applies.
Go deeper with our CCA-F prep course#
The exam guide tells you what's tested. Our Claude Certified Architect – Foundations prep course teaches you how to reason through it. Khayyam Hashmi, Educative's VP of Technical Content, who holds a PhD in computer science, built it around the exam blueprint.
Built on the blueprint. Every module targets one of the exam's five domains. It starts with the heaviest, agentic architecture, and builds from a single agent loop to multi-agent orchestration.
Practice in the exam's format. Each module ends with scenario-based questions written like the exam's, 43 quizzes in all. That matters more now that the official practice exam is gone.
Working code throughout. You build and run real implementations in the browser instead of reading about them.
A reliability capstone. The last module collects the anti-patterns the exam tests and walks through an end-to-end System Design and reliability audit.
Two lessons are free if you want to see the teaching style first: Probabilistic vs. Deterministic and Agents, Tools, and Context. An AI Tutor is available on every lesson when you get stuck.
If Professional is your goal, our Claude Certified Architect – Professional prep course covers its seven domains in 37 lessons and 35 quizzes, built around a running case study.
FAQ#
How many questions are on the Claude Certified Architect – Foundations exam?
There are 60 multiple-choice and multiple-response questions, answered in 120 minutes and grouped into four scenarios drawn from a bank of six.
What score do you need to pass?
You need a scaled score of 720 on a range of 100 to 1,000. The same bar applies to all four Claude certifications.
How much does the Claude Certified Architect exam cost?
Foundations costs $125 USD and Professional costs $175, before partner-tier discounts. Each retake costs the full fee.
Can I take the exam if my company isn't a Claude partner?
Not currently. Anthropic limits certification to people at Claude Partner Network organizations, who register with a recognized company email.
Do I need Foundations before taking Professional?
There's no formal prerequisite, although Anthropic recommends Foundations as the place to start.
Is there an official practice exam?
Anthropic retired it when exams moved to Pearson on June 30, 2026. The exam guide's sample questions are the official practice material now.
How long does the certification last?
It lasts 12 months from the date you earn it, and on-time renewal is a free, non-proctored assessment.
How long should I study?
That depends on your hands-on experience. The guide's typical candidate has six or more months of building with Claude, and with that background four focused weeks is realistic.
Where to go from here#
Start with the official exam guide and one of the builds above, then decide how much structure you want. If you'd like the prep mapped to the blueprint with practice in the exam's format, start with the two free lessons of our CCA-F prep course.