Claude Certified Architect (CCAR-F) by Anthropic
Free courses, exam guide and adaptive practice. Premium adds full mock exams and the AI tutor.
Claude Certified Architect – Foundations (official code CCAR-F, widely searched as CCA-F) is Anthropic's proctored exam for people who design production systems on the Claude API, Agent SDK, MCP and Claude Code. It costs $125, has 60 scored items in 120 minutes, passes at 720/1000, and stays valid for 12 months with free on-time renewal. You register through Pearson Professional Assessments.
PrepGenAICerts is an independent, third-party prep resource and is not affiliated with, endorsed by, or acting on behalf of Anthropic PBC.
Last updated October 20, 2026
It is an applied architecture exam, not a model-training or data-science test: every question is a real-world scenario asking you to choose the right design. This page explains what the program is, how it differs from cloud AI certifications, and how to prepare. For the full exam format, scoring, and domain deep-dives, see the complete CCA-F exam guide.
| Official name | Claude Certified Architect – Foundations |
|---|---|
| Exam code | CCAR-F / CCA-F |
| Fee (USD) | $125 |
| Scored items | 60 |
| Time | 120 minutes |
| Passing score | 720/1000 |
| Validity | 12 months |
| Renewal | To renew on time, review what has changed since certifying and complete a free, non-proctored assessment on the Anthropic Partner Academy. There is no fee for on-time renewal. |
| Delivery | Proctored: online proctored and/or test center, per program policy, delivered by Pearson VUE. |
| Register | Pearson Professional Assessments |
Source: Fees, item counts, and domain weights from Anthropic exam guides (v1.0, Effective July 2026) and the Partner Academy catalogue. Delivery and Partner Network eligibility from Pearson Professional Assessments (formerly Pearson VUE). Pearson Anthropic page · Partner Academy catalogue · Certification policies.
The five CCA-F exam domains
The exam blueprint is weighted across five technical domains. The percentages are the share of exam questions, so prioritise the heavier domains when you study.
Agentic Architecture & Orchestration
Design and implement agentic loops, multi-agent systems, coordinator-subagent patterns, hooks, task decomposition, and session management for production Claude applications.
Claude Code Configuration & Workflows
Configure CLAUDE.md hierarchies, create custom skills and commands, apply path-specific rules, choose plan mode vs direct execution, use iterative refinement, and integrate Claude Code into CI/CD.
Prompt Engineering & Structured Output
Design prompts with explicit criteria, apply few-shot patterns, enforce structured output via JSON schemas, implement validation loops, design batch processing strategies, and architect multi-instance reviews.
Tool Design & MCP Integration
Design effective tool interfaces, implement structured error responses, distribute tools across agents, integrate MCP servers, and use built-in Claude Code tools.
Context Management & Reliability
Manage conversation context across long interactions, design escalation and ambiguity resolution, implement error propagation, handle codebase exploration, design human review workflows, and preserve information provenance.
Who the CCA-F is for (and who it isn't)
Claude Certified Architect – Foundations (official code CCAR-F, searched as CCA-F) is built for people who design and implement production applications with Claude, not for people who only chat with a model. Anthropic's target profile is a solution architect with typically six or more months of practical work on the Claude API, the Agent SDK, Claude Code, and MCP: multi-agent orchestration, subagent delegation, tool integration, lifecycle hooks, CLAUDE.md and Agent Skills, MCP server integrations, plan mode, prompt design for reliable structured output, context-window management across long documents and handoffs, CI/CD integration, and escalation or reliability decisions.
That is an architecture exam. You are scored on design trade-offs: when a coordinator-plus-subagents pattern is justified versus a single agent; when stdio MCP is enough versus a remote transport; when to preserve full context versus summarise; when to escalate to a human. You do not need machine-learning research experience, model-training hours, or a data-science background. Software engineers and solution architects who ship Claude systems are the core audience.
It is not the right first exam if your work is using Claude in Claude.ai or Projects without writing integration code — that is Claude Certified Associate – Foundations (CCAO-F). It is also not the exam if you primarily write application code against the API, tools, and evals and do not own multi-agent production design — that is Claude Certified Developer – Foundations (CCDV-F). If you already hold CCA-F and design production systems end to end (integration, governance, evaluation, stakeholders), look at Claude Certified Architect – Professional (CCAR-P) instead of retaking Foundations. Compare codes and fees on the exam-codes page.
What are the CCA-F exam domains and how many items does each get?
The exam has 60 scored items in 120 minutes. Domain weights are the share of those items. Approximate item counts below are weight × 60, rounded, labelled approx. so you can plan study time. They are not a guarantee of the mix on any one form.
| Domain | Weight | Approx. items |
|---|---|---|
| Agentic Architecture & Orchestration | 27% | 16 |
| Claude Code Configuration & Workflows | 20% | 12 |
| Prompt Engineering & Structured Output | 20% | 12 |
| Tool Design & MCP Integration | 18% | 11 |
| Context Management & Reliability | 15% | 9 |
CCA-F (CCAR-F) domain weights and approximate scored items (weight × 60, rounded)
Sixteen of sixty items sit in Agentic Architecture. If you under-prepare that domain, you cannot make it up with perfect scores on Context Management. Study in weight order; use the concept library linked below rather than treating every domain as equal.
What each domain actually tests
Each domain is a cluster of design decisions, not a vocabulary quiz. The notes below are prose summaries of what the blueprint is asking you to judge. They are not exam items.
Agentic Architecture & Orchestration (27%)
This domain asks whether you can design an agentic system that will still be operable in production: loops, coordinator and subagent patterns, when to decompose a task, how agents share session state, and where SDK hooks belong in the lifecycle. Scenarios typically present two architectures that both "work" in a demo and ask which one fails under load, ambiguous goals, or a missing tool. Start with the Agentic Architecture concept library. Weight your week here first; it is the largest slice of the 60 items.
Claude Code Configuration & Workflows (20%)
You are tested on how teams actually run Claude Code: the CLAUDE.md hierarchy, Skills and custom commands, path-specific rules, plan mode versus direct execution, iterative refinement, and CI/CD. The recurring trap is putting enforcement in a markdown instruction when it belongs in settings.json, or leaving team conventions in a personal ~/.claude file so new hires never see them. Work through Claude Code Configuration & Workflows with a real repo in mind, not a toy prompt.
Prompt Engineering & Structured Output (20%)
The exam treats prompting as a reliability technique: explicit criteria, few-shot examples, JSON schemas and structured output, validation-plus-retry, batch strategies, and multi-instance review. You are not scored on clever wording. You are scored on whether the output can be parsed, checked, and retried without a human reading every token. The Prompt Engineering & Structured Output library is the matching study path. Pair it with glossary terms such as tool choice, JSON schema flags, and stop sequences when those appear in a design.
Tool Design & MCP Integration (18%)
This domain is interface design: tool names and schemas that a model can call reliably, structured errors the agent can recover from, how tools are distributed, when to use a built-in tool versus an MCP server, and transport trade-offs. Expect scenarios that look like "just add another tool" and are actually about blast radius, auth, or an error shape the orchestrator cannot parse. Study Tool Design & MCP Integration. Keep MCP facts in the glossary (mcp.json, Streamable HTTP, tools primitive) rather than mixing Developer-exam API trivia into Architect answers.
Context Management & Reliability (15%)
The smallest domain is still about production failure: what to keep in context, position effects, summarization risk, when to escalate on ambiguity, how errors propagate, graceful degradation, codebase exploration under a token budget, human-review workflows, and provenance across conflicting sources. A fluent answer that is missing a citation or that silently drops a constraint is a reliability miss. Use the Context Management & Reliability library. Fifteen percent is about nine items — enough to matter if you ignore it, not enough to rescue a weak Agentic Architecture score.
Exam day: format, delivery and scoring
CCA-F is a proctored exam delivered by Pearson VUE. You choose online proctoring (OnVUE) or a Pearson VUE test centre when you schedule. Bring a valid, unexpired, government-issued photo ID whose name matches the name on your registration exactly. Correct a name mismatch with certifications-support@anthropic.com before you schedule.
You get 60 scored items in 120 minutes. Items are scenario-based multiple-choice (and, where the form uses them, multiple-response). There is no live coding. The result is pass/fail with a scaled score from 100 to 1,000; 720 is the passing score. The score report also shows percent-correct by domain, which is the only official signal for a retake plan. The credential is valid for 12 months from the date it is awarded, with a free on-time renewal assessment on the Anthropic Partner Academy. If you do not pass, waiting periods are 14 days after the first attempt, 30 after the second, and 90 after the third, with a maximum of four attempts per exam per rolling year. Full mechanics sit on the shared policy reference and validity pages.
CCA-F vs CCAR-F: same exam
CCA-F and CCAR-F are the same Claude Certified Architect – Foundations exam. CCAR-F is the official code; CCA-F is the widely searched colloquial form. Do not register twice or study two blueprints. The mapping for all four credentials is on Claude certification exam codes.
Registration in 6 steps
Registration is the same six-step flow for every Anthropic Claude certification. The CCA-F fee at checkout is $125 USD (partner-tier discounts may apply). Detail for this credential lives on CCA-F cost and registration.
- Review the exam details on the Anthropic Partner Academy.
- Download the official Exam Guide and review the Certification Terms and Conditions and the Certification Exam Policy before registering.
- Register for the exam and complete checkout. The fee shown reflects any discount that applies to your partner tier.
- Follow the confirmation instructions to create your Pearson VUE account, then sign in to schedule.
- Choose an available date and select either online proctoring or a Pearson VUE test center.
- You may cancel or reschedule up to 24 hours before your appointment. Changes made within 24 hours forfeit the exam fee.
Anthropic CCA-F vs Google Cloud & AWS AI certifications
The CCA-F is often compared to cloud machine-learning certifications, but it tests a different skill set. Cloud ML certs validate training and operating models; the CCA-F validates building applications on top of a frontier model. If your work is building with Claude rather than training models, the CCA-F is the directly relevant credential.
| Certification | Focus | What it tests | Audience |
|---|---|---|---|
| Anthropic CCA-F | Building applications with a frontier LLM (Claude) | Agentic architecture, tool & MCP design, prompt engineering, Claude Code, context management | Software engineers & solution architects building with Claude |
| Google Cloud — Professional ML Engineer | Designing & operating ML models on Google Cloud | Model training, ML pipelines, Vertex AI, MLOps, responsible AI | ML engineers on Google Cloud |
| AWS — Machine Learning | Building & deploying ML on AWS | Data engineering, modeling, SageMaker, ML operations on AWS | ML practitioners on AWS |
How to prepare for the Anthropic certification
A focused candidate with hands-on Claude experience can prepare in two to four weeks. Every resource below is free to start.
Learn the domains
Work through the five domains in the concept library, weighting time by exam weight (Agentic Architecture is the heaviest at 27%). Use the glossary for any unfamiliar term.
Open the concept library →Practice with mock exams
Take scenario-based mock exams that mirror the real exam's format. Treat each as a diagnostic — note which domains drop below 70%.
Go to mock exams →Review weak areas & strategy
Use spaced-repetition review to reinforce low-accuracy concepts, then run the strategy and anti-pattern checklist before exam day.
Read exam strategy →Anthropic Certification — Frequently Asked Questions
Is there an official Anthropic certification?
Yes. Anthropic's certification program includes the Claude Certified Architect — Foundations (CCAR-F / CCA-F), an official credential for developers and architects who build production systems with the Claude API, Agent SDK, Model Context Protocol (MCP), and Claude Code. The exam and its five-domain blueprint are defined by Anthropic. Registration is open to people at organizations in the Claude Partner Network (company-domain email required); scheduling runs through the Anthropic Partner Academy and Pearson.
Is “Anthropic certification” the same as “Claude certification”?
They refer to the same thing. People search for both “anthropic certification” and “claude certification,” but the actual credential is named the Claude Certified Architect (CCA-F) — Anthropic is the company, Claude is the model family the exam covers.
How does the Anthropic CCA-F compare to Google Cloud or AWS AI certifications?
Google Cloud's Professional Machine Learning Engineer and AWS's Machine Learning certifications validate building and operating ML models on a cloud platform. The CCA-F is different in kind: it validates designing applications on top of a frontier LLM — agentic loops, tool and MCP design, prompt engineering, Claude Code workflows, and context management. If your work is building with Claude rather than training models, the CCA-F is the directly relevant credential.
Do I need machine-learning experience to take the CCA-F?
No. Unlike cloud ML certifications, the CCA-F does not test model training, MLOps, or data-science math. It tests application-architecture judgment when building with Claude — so software engineers and solution architects are the core audience, not ML researchers.
How long does it take to prepare for the Anthropic certification?
Most candidates with hands-on Claude experience prepare in two to four weeks of focused study. A typical path is one week per domain cluster using the free concept library and glossary, then mock exams and spaced-repetition review in the final week. See the preparation timeline above.
Does the CCA-F certification have a validity or expiry period?
Yes. The Claude Certified Architect – Foundations (CCAR-F / CCA-F) credential is valid for 12 months from the date it is awarded. To renew on time, review what has changed since certifying and complete a free, non-proctored assessment on the Anthropic Partner Academy. There is no fee for on-time renewal. If the credential lapses, if a credential lapses, the holder must retake the full exam at the full fee to regain certified status. Full details: /claude-certified-architect/validity.
Everything about the CCA-F certification
CCA-F prep course: free lessons
Free structured lessons covering all five exam domains.
CCA-F cheat sheet
Printable quick-reference cards for every CCA-F exam domain.
Benefits
Why the credential is worth earning and what it signals to employers.
Eligibility
Who this exam is for and the experience Anthropic recommends.
Validity & Renewal
How long the credential lasts and how to keep it current.
Cost & Registration
Exam fee, the 6-step registration process, and scheduling.
Frequently Asked Questions
20 questions candidates actually ask, answered in depth.
Compare all 4 Claude certifications
See how the CCA-F compares to the Professional, Associate, and Developer credentials.
See the other Claude certifications
Start preparing for the CCA-F — free
Concept library, glossary, and mock exams across all five domains. No account required to start learning.