CCA-F Exam Overview

The CCA-F exam — full name Claude Certified Architect — Foundations — is Anthropic's first publicly available certification for engineers and architects who design production systems with Claude. This study note is the orientation page for anyone who has just heard of the certification and needs the facts before deciding whether to sit it. Everything below is sourced from the official Anthropic CCA-F Exam Guide (v0.1, February 2025) and the public Anthropic certification announcement; nothing here is rumour, NDA-protected content, or extrapolation. If you came here to find out whether the CCA-F exam is a fit for your career, your team, or your roadmap, this page should answer the question in one read.

The CCA-F exam sits in a new category of vendor certifications: it does not test knowledge of an SDK in the way an AWS or Azure exam does, and it does not test ML theory in the way a Google ML certification does. It tests architectural decision-making for systems where Claude is the reasoning engine — the choices a senior engineer makes when designing agents, integrating tools, configuring Claude Code, and managing context across long-running interactions. If you have ever shipped a Claude-powered feature to production and made a call about agentic loops, MCP servers, or coordinator-subagent patterns, the CCA-F exam is testing exactly that decision surface.

What Is the CCA-F Exam?

The CCA-F exam is the foundational tier of the Claude Certified Architect track. Anthropic launched it on March 12, 2026 as the first generally available exam in a multi-tier certification program (sellers, developers, and advanced architect tiers are scheduled later in 2026). The exam credential validates that a candidate can make sound architectural decisions for production Claude systems across five domains: agentic architecture and orchestration, tool design and MCP integration, Claude Code configuration and workflows, prompt engineering and structured output, and context management and reliability.

The CCA-F exam is delivered online through Anthropic's Skilljar learning platform. It is not a hands-on lab exam. Every question is a written scenario followed by multiple-choice options. The credential, once earned, is valid indefinitely; Anthropic has not published a recertification cadence as of the launch window.

CCA-F Exam Format — 60 Questions, 120 Minutes, 720 Passing Score

The CCA-F exam format is fixed and published in the official exam guide. There is no adaptive-difficulty mechanic, no penalty for wrong answers, and no partial credit.

A candidate has on average two minutes per question. That is generous compared to most cloud architect certifications, but every CCA-F question is a scenario stem (often 4–8 sentences) followed by 4 options of 1–3 lines each, so the time pressure is roughly equivalent to AWS Solutions Architect Associate. Most pass-reporters in the early launch window finished with 15–30 minutes to spare and used the leftover time to review flagged questions.

The Six CCA-F Scenarios — How Anthropic Tests Architectural Decisions

The most distinctive element of the CCA-F exam — and the one most often misunderstood by first-time candidates — is the scenario rotation. Anthropic frames every architectural concept inside one of six published scenarios. On any given sitting, four of the six scenarios are randomly selected and questions are drawn from those four; the other two scenarios are silent on your test form.

The six CCA-F scenarios are:

1. Customer Support Resolution Agent — autonomous ticket handling with knowledge-base lookup, CRM queries, and human escalation.
2. Code Generation with Claude Code — Claude Code as a code authoring and refactoring assistant inside a real codebase.
3. Multi-Agent Research System — coordinator-subagent fan-out for parallel research tasks with result aggregation.
4. Developer Productivity with Claude — SDK-driven agents performing autonomous codebase tasks (bug investigation, refactors, test authoring).
5. Claude Code for Continuous Integration — non-interactive Claude Code invocation in CI/CD pipelines for automated PR review and test generation.
6. Structured Data Extraction — extracting structured records from unstructured documents with validation and human review.

Each scenario carries the same conceptual surface in different clothing. A question about agentic loops is presented as a customer-support ticket investigation in one scenario and a multi-file refactor in another, but the underlying answer turns on the same stop_reason mechanics. Recognising that the CCA-F scenarios are wrappers around the same five domains — not six independent topics — is one of the bigger early aha moments.

Why 4 of 6, Not All 6?

Anthropic deliberately rotates scenarios so that a candidate cannot memorise scenario-specific answers. You must be architecturally fluent across all six because you do not know which four you will see. This is the single most important framing for CCA-F exam preparation: study every scenario, expect to meet four, and treat any scenario-specific shortcut as unreliable.

CCA-F Five Domains and Their Weights

The CCA-F exam scores you across five published domains. Domain weights determine how many questions are drawn from each topic area; a 60-question exam at the published weights produces the question counts in the rightmost column.

Domain 1 (agentic architecture) is the single largest weight, and inside Domain 1 the agentic-loop and coordinator-subagent topics account for the majority. If you have to cut study time anywhere, do not cut Domain 1. Domains 2 through 5 are roughly equal-importance.

CCA-F Question Types — Multiple Choice with Single Correct Answer

The CCA-F exam uses one question type only: single-correct multiple choice. Each question presents:

• A scenario stem describing a system, a problem, or an architectural decision.
• Four answer options labelled A, B, C, D.
• Exactly one correct answer.

There are no multiple-response questions ("select all that apply"), no ordering or sequencing questions, no drag-and-drop, no code-fill-in-the-blank, and no lab tasks. This is a simpler item format than AWS Solutions Architect or Google Professional Cloud Architect exams, and it makes time management more predictable.

The single-correct format does, however, mean that distractor design is sharp. Anthropic frequently writes options that are partially correct or correct in a different scenario — and you have to pick the best fit for the given stem. Reading the stem carefully (especially the verbs and adjectives — "autonomous", "scripted", "real-time", "batch") is more important on CCA-F than on most cert exams.

CCA-F Scoring Methodology — Scaled, Not Raw Percentage

The CCA-F exam uses a scaled scoring model with the following published parameters:

• Score range: 0 to 1000
• Passing score: 720
• All 60 questions are scored (no unscored pretest items announced)
• No partial credit on multi-block questions (since there are none)
• Negative marking: none (wrong answers do not subtract points; always answer every question)

A scaled score of 720 corresponds approximately — but not exactly — to a raw correct rate of 70–75%. The scaling reflects question difficulty: missing one hard question hurts less than missing one easy question. Anthropic does not publish per-question scoring weights or domain-level pass requirements.

CCA-F Prerequisites and Recommended Background

There are no formal prerequisites for the CCA-F exam. Anyone can register and sit it without prior certifications, employer affiliation, or Anthropic Partner Network membership.

That said, Anthropic recommends candidates have:

• 6+ months of hands-on experience building with Claude (API, Agent SDK, or Claude Code).
• Working familiarity with agentic patterns — building at least one agent that loops on tool_use is highly recommended.
• Comfort with JSON Schema for structured output enforcement.
• Working knowledge of Model Context Protocol (MCP) at the integration level (you do not need to operate MCP servers, but you should know what they are and when to use them).
• General software engineering background — production system design, observability, error handling.

Candidates from a pure prompt-engineering background (no agent-building experience) tend to score lower on Domain 1 because the scenario depth assumes you have wrestled with stop_reason mechanics, parallel tool calls, and termination guards in code. Candidates from a pure DevOps/infrastructure background tend to score lower on Domain 4 (prompt engineering and structured output) for the symmetric reason. The exam rewards candidates who have shipped at least one Claude-powered feature end-to-end.

CCA-F Cost, Booking, and Delivery Method

The published price for the CCA-F exam is $99 USD for standard candidates. Members of the Claude Partner Network can take it for free, subject to Anthropic's partner program terms.

Booking happens through Anthropic's Skilljar instance:

• Booking URL: https://anthropic.skilljar.com/claude-certified-architect-foundations-access-request
• Delivery: online, proctored
• Identity verification: government photo ID required during proctor check-in
• Environment: quiet room, single screen, no second monitor, no notes
• Browser: standard online proctoring constraints (clean room, no other applications)

Once you complete the access request, scheduling and proctor session details are issued by Skilljar. Anthropic does not currently offer in-person testing centres for CCA-F.

CCA-F Retake Policy and Score Reports

Anthropic's published retake policy at the CCA-F exam launch window:

• If you fail, you may retake the exam after a waiting period (specific cooling-off interval published in the candidate handbook on Skilljar).
• Each attempt is paid separately at the same $99 standard price; Partner Network free attempts are subject to program terms.
• You receive a pass/fail result and a scaled score at the end of the exam.
• Domain-level breakdown is provided in the post-exam score report; this lets you see which domains pulled your score down for retake planning.
• Specific question-level feedback is not provided (this is standard for proctored cert exams).

If you fail by a small margin (e.g. 700–719), Anthropic's score report typically shows you which one or two domains underperformed; targeted study of those domains for two to three weeks before a retake is the most efficient path back.

Plain-Language Explanation: CCA-F at a Glance

Three analogies anchor the CCA-F exam for candidates who want a fast intuitive map before drilling into the formal content.

Analogy 1: The Driver's Licence Theory Test

Think of the CCA-F exam as the theory portion of a driver's licence test. A driver's licence theory test does not put you behind the wheel; it asks whether you understand the rules of the road, the traffic signs, and the right-of-way decisions a competent driver makes. CCA-F is the same shape: it does not put you in front of an SDK. It asks whether you understand the rules of agent design, the meaning of stop_reason, and the right architectural call when six tools could each plausibly do the job. Just as a theory test focuses on decisions, not mechanical skill, CCA-F focuses on architectural decisions, not coding fluency. You can pass the theory test without being a great driver, and you can pass CCA-F without being the fastest typist — but only if you genuinely understand the rules.

Analogy 2: The Open-Book Exam That Closes Right at the Door

The CCA-F exam is a closed-book exam, but it tests material that in real life you would always look up — the exact stop_reason values, the precise AgentDefinition schema fields, the canonical MCP resource shape. Treat it like an open-book exam where the proctor closes the book at the door. In your real job, you can Cmd+F the docs in 5 seconds. On exam day, you must carry the docs in your head. The implication for study: do not read the documentation passively. Make flashcards for every API field, every config option, every CLI flag that the official Anthropic guide marks as testable. The test does not reward "I know where to find this." It rewards "I have memorised this."

Analogy 3: The Restaurant Tasting Menu — Six Courses, Four Served

The CCA-F scenarios work like a tasting menu where six courses are prepared and only four are served at random. The chef (Anthropic) commits to the menu in advance; you commit to studying every course. On exam day, the kitchen rolls dice and four plates arrive. You cannot ask for a menu reveal; you cannot send a course back; you cannot trade a scenario you do not like for one you do. The implication for study: do not pre-commit time only to your strongest scenarios. The two scenarios you skip might be exactly the two that arrive on your test form. Spend equal time on each of the six CCA-F scenarios and you will be calm regardless of which four show up.

CCA-F Comparison with Other AI / Cloud Architect Certifications

The CCA-F exam is lighter than most cloud architect certs in scope (one vendor, one model family) but deeper in agent-design surface area than any of them. It is also the cheapest. If you already hold a cloud architect cert, you will find the CCA-F format familiar but the content unfamiliar — agentic loops and MCP have no analogue on the AWS/GCP/Azure exams.

What CCA-F Is NOT — Out of Scope Topics

The official CCA-F exam guide publishes an explicit Appendix listing what the exam will not test. This list is gold for study planning — it tells you what to skip.

• Fine-tuning or training custom Claude models
• API authentication, billing, or account management details
• Specific programming language / framework implementation details
• MCP server deployment / hosting infrastructure (networking, containers, k8s)
• Claude's internal architecture, training process, or model weights
• Constitutional AI, RLHF, or safety training methodologies
• Embedding model or vector database implementation details
• Computer use (browser automation, desktop interaction)
• Vision / image analysis capabilities
• Streaming API implementation details (server-sent events, WebSockets)
• Rate limiting, quotas, or API pricing calculations
• OAuth, API key rotation, or authentication protocol details
• Specific cloud provider configurations (AWS Bedrock, Azure, GCP Vertex)
• Performance benchmarking or model comparison metrics
• Prompt caching implementation details (beyond knowing the feature exists)
• Token counting algorithms or tokenization specifics

If you find study material in any of the above areas, set it aside for now. It will not appear on the CCA-F exam. Anthropic has signalled that some of these areas (vision, computer use) will appear in the advanced architect tier scheduled for later 2026.

Common Misconceptions Newcomers Have About CCA-F

Six recurring misconceptions show up in early-window community discussion. Clearing them now saves study time.

Misconception 1: "CCA-F is a coding exam."

It is not. There is no code editor. There are no fill-in-the-blank questions. Every item is a scenario stem with multiple-choice options. You will be tested on architectural decisions, not on whether you can write Python or TypeScript that calls the Messages API.

Misconception 2: "If I know the Anthropic docs cold, I will pass."

Necessary but not sufficient. The CCA-F exam tests applied judgment — picking between two architecturally plausible patterns based on scenario constraints. Knowing the docs gets you to recognising the options; you still need to recognise which is the best fit for the given scenario.

Misconception 3: "I should only study the scenarios most relevant to my job."

Risky. Because 4 of 6 scenarios rotate randomly, skipping any scenario is a one-in-three chance of missing a quarter of the test. Study breadth across all six scenarios.

Misconception 4: "CCA-F is the only Claude cert I need."

CCA-F is the foundational tier. Anthropic has announced advanced architect, developer, and seller tiers for later 2026. CCA-F is your starting point, not your endpoint.

Misconception 5: "I should memorise example code from the docs."

Memorising long code blocks is high-cost / low-yield. The exam asks about shapes (the four stop_reason values, the AgentDefinition fields, the tool_result block structure), not full implementations. Make a one-page cheat sheet of shapes; do not transcribe entire code blocks.

Misconception 6: "Pass rate is published."

Anthropic does not publish official pass rates for CCA-F. Any specific percentage you see online is community estimate, not Anthropic data. Treat such numbers with healthy skepticism.

FAQ — CCA-F Exam Format Top Questions

How long is the CCA-F exam and how many questions are there?

The CCA-F exam is 120 minutes long and contains 60 multiple-choice questions. That is an average of 2 minutes per question. Most successful candidates finish with 15–30 minutes of buffer time and use the remainder to review questions they flagged for second look. The exam is delivered online via Skilljar with a single proctored session.

What is the passing score for CCA-F?

The CCA-F passing score is 720 out of 1000 on a scaled score scale. This corresponds approximately to 70–75% raw correct depending on the difficulty mix of the questions you receive, but Anthropic does not publish exact raw-score thresholds. Aim for high-70s on practice exams before booking the real CCA-F exam to give yourself comfortable headroom.

How much does the CCA-F exam cost?

The standard price is $99 USD. Members of the Anthropic Claude Partner Network can take the CCA-F exam for free, subject to partner program terms. Each retake is paid separately at the same standard price (or free for partners). There are no published bulk-discount or enterprise pricing tiers at the launch window.

What scenarios are tested on CCA-F and how many appear on my test form?

There are six published CCA-F scenarios: customer-support-resolution-agent, code-generation-with-claude-code, multi-agent-research-system, developer-productivity-with-claude, claude-code-for-continuous-integration, and structured-data-extraction. Four of the six are randomly selected for each test form. You cannot choose which four; you cannot trade scenarios; you must prepare all six. Each scenario carries roughly 25% of the per-form question weight, so missing any scenario is significant.

What are the prerequisites for taking the CCA-F exam?

There are no formal prerequisites. Anthropic recommends 6+ months of hands-on Claude experience, comfort with agentic patterns, working knowledge of MCP, and general software engineering background. You do not need to be a Partner Network member, hold any prior cert, or work for an Anthropic customer. Anyone can register and pay $99.

What is out of scope on the CCA-F exam?

A long list, published in the official guide Appendix: fine-tuning, RLHF, Constitutional AI, vision/computer use, embedding internals, MCP infrastructure deployment, OAuth/auth protocol details, cloud provider specifics (Bedrock/Vertex/Azure), token-counting internals, streaming implementation, and rate-limit/pricing calculations. If you find study material in any of these, set it aside — it will not appear on CCA-F. Some of these (vision, computer use) are planned for the advanced tier scheduled later in 2026.

Can I take the CCA-F exam in languages other than English?

At the launch window, CCA-F is delivered in English only. Anthropic has not published a localisation roadmap. If English is not your first language, plan for slightly more time per question — the exam is still finishable in 120 minutes for most candidates regardless of native language, but the dense scenario stems benefit from extra reading time.

Is there a recertification cycle for CCA-F?

Anthropic has not published a recertification cadence at the launch window. The CCA-F credential, once earned, appears valid indefinitely. Given that Claude evolves rapidly, expect Anthropic to publish a refresh policy in the future; check the official certification page periodically after passing.

Further Reading — Official Anthropic Sources

• Official CCA-F Exam Guide (PDF, v0.1 Feb 2025): https://everpath-course-content.s3-accelerate.amazonaws.com/instructor/8lsy243ftffjjy1cx9lm3o2bw/public/1773274827/Claude+Certified+Architect+%E2%80%93+Foundations+Certification+Exam+Guide.pdf
• CCA-F booking and access request: https://anthropic.skilljar.com/claude-certified-architect-foundations-access-request
• Building effective agents — agentic loop reference: https://docs.anthropic.com/en/docs/build-with-claude/agentic-loop
• Agent SDK overview: https://docs.anthropic.com/en/docs/claude-code/sdk/sdk-overview
• Tool use overview: https://docs.anthropic.com/en/docs/agents-and-tools/tool-use/overview
• Anthropic certifications announcement: https://www.anthropic.com/news/claude-certifications

Related ExamLab notes: CCA-F Exam Tips and Strategies, Agentic Loops for Autonomous Task Execution, Multi-Agent Orchestration with Coordinator-Subagent Patterns.