Claude Certified Architect (CCAR-F) Exam Guide 2026

Claude Certified Architect – Foundations (CCAR-F) is Anthropic's foundational architect credential for professionals who design production Claude solutions. The exam contains 60 questions, allows 120 minutes, uses a scaled passing score of 720 out of 1,000, costs $125 USD, and remains valid for 12 months. It focuses on agentic architecture, Claude Code workflows, prompt engineering, tool and MCP design, and context reliability.

Last reviewed: August 22, 2026. Exam details can change, so confirm current policies on the official Anthropic certification page before registering.

What Is the Claude Certified Architect Foundations Certification?

The Claude Certified Architect Foundations certification validates whether you can make sound design decisions for real Claude-based systems. It is aimed at solution architects, AI architects, technical leads, senior developers, consultants, and engineers who need to choose appropriate models, architecture patterns, tools, context strategies, evaluation methods, and safety controls.

This is not a general artificial intelligence theory exam. It tests practical judgment: when to use a workflow instead of an autonomous agent, how to structure tools and MCP integrations, where Claude Code configuration belongs, how to produce reliable structured output, and how to manage long-running context without losing critical information.

Anthropic launched the credential as part of the Claude Partner Network. The program now includes separate Associate, Developer, Architect Foundations, and Architect Professional credentials, each aligned to a different role.

Who Can Register for CCAR-F?

Anthropic currently makes its certification exams available to people whose organizations participate in the Claude Partner Network. Registration requires a recognized company email address; a personal email address does not satisfy the current access requirement.

There is no lower-level certification prerequisite for Architect Foundations. However, this is a technical, scenario-based exam, so candidates benefit from hands-on familiarity with the Claude API, Claude Code, agent workflows, tool use, and Model Context Protocol. Treat practical experience as preparation rather than as a formal prerequisite unless Anthropic's current registration page says otherwise.

CCAR-F Exam Format and Scoring

The exam gives you 120 minutes for 60 questions, which works out to an average of two minutes per question. Questions emphasize architectural choices and production tradeoffs rather than memorized definitions. Expect both single-answer multiple-choice items and multiple-response items that tell you how many answers to select.

The passing score is 720 on a scaled range from 100 to 1,000. A scaled score is not the same as answering exactly 72% of the questions correctly. Question difficulty and the scoring model determine the final result, so prepare to perform consistently across all five domains rather than relying on one strong topic.

The Five CCAR-F Exam Domains

The official blueprint divides the Claude Architect Foundations exam into five weighted domains. Use the percentages to allocate study time, but do not ignore the smaller domains: the exam assesses whether you can combine these skills in realistic solution scenarios.

1. Agentic Architecture and Orchestration – 27%

This is the largest domain. Focus on deciding between deterministic workflows and model-directed agents, designing orchestrator-worker and multi-agent patterns, managing tool loops, enforcing limits, preserving state, and introducing human review where risk requires it.

You should be able to identify when agent autonomy improves a solution and when it creates unnecessary cost, latency, nondeterminism, or operational risk.

2. Claude Code Configuration and Workflows – 20%

Study the scope and purpose of project and user instructions, CLAUDE.md files, commands, hooks, permissions, skills, headless operation, and CI/CD integration. A common decision pattern is choosing the narrowest configuration scope that reliably reaches the intended users without loading irrelevant guidance into every interaction.

3. Prompt Engineering and Structured Output – 20%

This domain covers system prompts, examples, XML organization, output constraints, validation, retry strategies, and explicit decision boundaries. The exam rewards prompts that define positive, negative, and ambiguous cases instead of forcing Claude to guess.

Know how to combine prompt instructions with schemas and deterministic application validation. A valid JSON response can still be factually unsupported, so semantic boundaries matter as much as syntax.

4. Tool Design and MCP Integration – 18%

Prepare to design tools with clear names, descriptions, input schemas, output contracts, permissions, and failure behavior. Understand the difference between MCP tools, resources, and prompts, and be able to select sensible server boundaries and transports.

When several tools overlap, clearer purpose-specific interfaces usually outperform one large tool with many optional modes. The exam tests whether your design helps Claude select and use capabilities safely.

5. Context Management and Reliability – 15%

This domain tests context-window management, compaction, retrieval, prompt caching, durable state, escalation, confidence handling, and human-in-the-loop review. You should know how to preserve exact facts and provenance while removing verbose intermediate material that no longer belongs in active context.

Reliable systems separate durable state from transient conversation. Structured records, targeted retrieval, and explicit unresolved fields are often safer than replacing an entire history with an unverified prose summary.

What the CCAR-F Exam Really Tests

Strong candidates can explain not only how a Claude feature works, but why one architecture is safer or more efficient than another. Expect decisions involving:

• Workflow versus agent: whether the task follows known steps or requires adaptive model-directed decisions.

• Host-controlled safety: enforcing tool-call limits, permissions, timeouts, validation, and escalation outside the model.

• Tool boundaries: giving Claude distinct, understandable capabilities instead of ambiguous interfaces.

• Context design: preserving critical facts, citations, and unresolved work without sending every prior token.

• Structured output: combining clear semantic instructions, schemas, validation, and safe handling of uncertainty.

• Cost and latency: choosing models, caching, batching, and architecture patterns appropriate to the workload.

• Evaluation: defining success criteria, testing decision boundaries, and monitoring production behavior.

CCAR-F vs CCDV-F: Which Claude Certification Fits You?

Comparison	CCAR-F: Architect Foundations	CCDV-F: Developer Foundations
Best for	Architects, technical leads, consultants, senior engineers	Developers building Claude applications and agents
Primary focus	End-to-end solution design and architectural tradeoffs	Implementation with the Claude API, tools, MCP, and Claude Code
Typical question	Which architecture best balances reliability, safety, cost, and context?	Which implementation correctly uses a Claude capability?
Prerequisite relationship	CCDV-F is not required	CCAR-F is not required

Choose CCAR-F if you regularly design systems, evaluate architecture options, set production boundaries, or advise stakeholders. Choose CCDV-F if your work is more implementation-focused. For a comparison of the complete credential family, read the guide to all four Claude certifications.

A Practical Two-to-Four-Week CCAR-F Study Plan

Week 1: Learn the Blueprint and Establish a Baseline

• Read the official exam guide and turn each objective into a study checklist.

• Take a short diagnostic across all five domains.

• Review the Claude API, Claude Agent SDK, Claude Code, and MCP fundamentals.

• Spend extra time on Agentic Architecture and Orchestration because it represents 27% of the exam.

Weeks 2 and 3: Build and Practice by Domain

• Build a small tool-using agent with a host-enforced iteration limit.

• Configure project-level Claude Code instructions, a reusable command, and a validation hook.

• Create an MCP server with clearly separated tools and explicit failure responses.

• Practice domain questions and review why every incorrect option fails the scenario.

• Keep a mistake log organized by domain and architecture principle.

Week 4: Timed Exam Preparation

• Complete at least two 60-question simulations under the 120-minute limit.

• Practice identifying the requirement that differentiates two plausible answers.

• Revisit weak domains and retake focused question sets.

• Confirm registration, identification, system, and testing-center requirements before exam day.

Common CCAR-F Preparation Mistakes

• Memorizing terms without comparing designs: the exam rewards architectural judgment, not isolated definitions.

• Ignoring host-side controls: production limits, permissions, validation, and escalation should not depend only on model compliance.

• Studying every domain equally: use the official weights while still maintaining minimum competence across all five domains.

• Forcing an answer when evidence is missing: reliable systems preserve uncertainty and unresolved fields rather than inventing a value.

• Using only untimed questions: full simulations reveal pacing problems and decision fatigue that domain drills cannot show.

Frequently Asked Questions

What is the Claude Certified Architect Foundations exam?

CCAR-F is Anthropic's foundational architect certification. It validates your ability to design reliable Claude-based solutions involving agents, Claude Code, prompt engineering, tools, MCP, context management, and production safeguards.

How much does the CCAR-F exam cost?

The listed exam price is $125 USD before any eligible Claude Partner Network discount. Anthropic increased the Architect Foundations price from $99 to $125 effective June 30, 2026.

How many questions are on the CCAR-F exam?

The exam has 60 questions and allows 120 minutes. That gives you an average of two minutes per question.

What score do I need to pass?

You need a scaled score of 720 or higher on a 100–1,000 scale. A scaled score is not a direct percentage of questions answered correctly.

Who is eligible to take the exam?

Anthropic currently limits certification registration to people at organizations in the Claude Partner Network. Candidates must register with a recognized company email address.

Is CCDV-F required before CCAR-F?

No. Claude Certified Developer Foundations can improve implementation fluency, but it is not a prerequisite for Claude Certified Architect Foundations.

How long is the certification valid?

The Claude Certified Architect Foundations certification is valid for 12 months. Check Anthropic's current renewal guidance as your expiration date approaches.

What is the best way to prepare?

Begin with the official exam guide and Anthropic Academy material, build small Claude systems using agents, tools, MCP, and Claude Code, then practice scenario-based questions across all five weighted domains. Finish with full-length timed simulations.

Official Resources and Next Steps

• Official Claude Certified Architect Foundations page

• Official Claude certification FAQ

• Free CCAR-F practice test and domain questions

• Claude Architect vs AWS AI Practitioner

The Claude Architect path is a strong fit when your work involves designing complete Claude solutions rather than only implementing individual API calls. Use the official blueprint to guide your study, build hands-on systems, and verify your readiness with domain practice and timed simulations before scheduling the exam.