What Is on the Claude Certified Architect Exams? Foundations vs Professional Domains, Format and Prep (2026)

Claude Certified Architect – Foundations tests whether you can build production agent systems with the Claude Agent SDK, Claude Code, the Claude API and MCP. Claude Certified Architect – Professional tests whether you can architect, integrate, evaluate and govern Claude in an enterprise. Foundations has five domains and 60 questions. Professional has seven domains and 63 questions. Both are 120 minutes and pass at 720 out of 1,000.

Everything in this guide comes from Anthropic's own exam guides for each exam (version 1.0, effective July 2026), linked in the sources. Where something is not in an official document, I say so.

This post is about what is on the exams. If you are still deciding whether to sit one, read is the Claude Certified Architect worth it. If you are choosing between Architect, Developer and Associate, read Architect vs Developer vs Associate.

Exam update on 7 December 2026. Pearson VUE says Architect – Professional will include updated content from 7 December 2026. Architect – Foundations is not part of that update. Anthropic has not yet published what changes. This guide reflects the current exam guides (v1.0, July 2026) and will be updated when the new guides are out.

Exam format: Foundations vs Professional

The two exams share the same delivery, scoring and validity rules. What differs is the price, the question count, and how the questions are structured.

A few details from the official guides that change how you prepare:

• Each multiple-response question tells you how many answers to pick. No guessing whether it is two or three.
• Your score report shows percent correct by domain. That only informs you; the pass or fail decision comes from your total scaled score.
• It is closed-book. No notes, no second monitor, no Claude. You reason about Claude architectures without Claude.
• Partner tier discounts apply automatically at checkout. Select, Preferred and Global Premier partners get 50% off, and Global Premier partners get 100% off through 31 December 2026.
• You can cancel or reschedule up to 24 hours before. Inside 24 hours, you lose the fee.

Retakes are strict: waits of 14, 30 and then 90 days, a maximum of four attempts per exam in a rolling 12 months, and full price each time. Total seat time is around 135 minutes once check-in is included. Full cost breakdown in Claude certification exams and costs.

Who can sit these exams

Only people at Claude Partner Network organisations, using a company-domain email. Personal addresses are rejected.

Anthropic launched the Claude Partner Network on 12 March 2026 with Architect – Foundations as its first certification. Professional followed in July 2026. Both remain partner-only as of today.

If you are on a Gmail or Outlook address, read can you take the Claude exams without a company email before you spend anything on prep. If your employer is a partner, how to register for Claude certification walks through booking.

Architect – Foundations domains (CCAR-F)

Foundations has five domains, and agentic architecture is the biggest single block at 27%. These are the official weightings from Anthropic's exam guide:

Anthropic describes the ideal candidate as a solution architect with 6+ months of practical experience building with the Claude API, Agent SDK, Claude Code and MCP.

Domain 1: Agentic Architecture & Orchestration (27%)

This domain tests whether you can design agent loops and multi-agent systems that behave predictably. The official task statements cover:

• Agentic loops: continuing on stop_reason "tool_use" and stopping on "end_turn", not parsing text or relying on arbitrary iteration caps.
• Coordinator and subagent patterns: hub-and-spoke design, and the fact that subagents do not inherit the coordinator's context automatically.
• Spawning subagents with the Task tool, passing context explicitly, and running subagents in parallel.
• Enforcement and handoff: when to use hooks and programmatic gates instead of prompt instructions, for example blocking a refund until a customer is verified.
• Agent SDK hooks such as PostToolUse for normalising data and blocking policy-violating tool calls.
• Task decomposition: fixed prompt chains versus adaptive decomposition.
• Session state: resuming named sessions, forking, and when to start fresh with a summary instead.

The recurring theme: when the business rule must hold every time, enforce it in code, not in the prompt.

Domain 2: Tool Design & MCP Integration (18%)

This domain tests whether you can give an agent the right tools, described well enough that it picks the right one. It covers:

• Writing tool descriptions that separate similar tools clearly, and renaming or splitting tools that overlap.
• Structured MCP error responses with the isError flag, error categories and a retryable flag.
• Giving each agent only the tools its role needs, and tool_choice options: "auto", "any" and forcing a specific tool.
• MCP server scope: project-level .mcp.json for the team versus user-level ~/.claude.json for personal servers, with environment variables for secrets.
• Choosing the right built-in tool: Grep for content, Glob for file paths, Edit for targeted changes, Read and Write as the fallback.

Domain 3: Claude Code Configuration & Workflows (20%)

This domain tests whether you can set Claude Code up for a team, not just for yourself. It covers:

• The CLAUDE.md hierarchy (user, project, directory), @import, and .claude/rules/.
• Custom slash commands in .claude/commands/ and skills in .claude/skills/ with frontmatter such as context: fork and allowed-tools.
• Path-scoped rules using glob patterns, for conventions like test files spread across a codebase.
• Plan mode versus direct execution.
• Iterative refinement: concrete examples, test-driven iteration and the interview pattern.
• Running Claude Code in CI/CD with -p, --output-format json and --json-schema.

Domain 4: Prompt Engineering & Structured Output (20%)

This domain tests whether you can get reliable, machine-readable output with few false positives. It covers:

• Explicit criteria instead of vague instructions like "be conservative".
• Few-shot examples for ambiguous cases and consistent formatting.
• Structured output with tool use and JSON schemas, including nullable fields so the model does not invent missing values.
• Validation and retry loops, and recognising when a retry cannot help because the information is not in the source.
• The Message Batches API: when its cost saving fits and when its latency does not.
• Multi-pass review with an independent instance instead of self-review.

Domain 5: Context Management & Reliability (15%)

This domain tests whether your system stays accurate over long sessions and fails safely. It covers:

• Keeping critical facts (amounts, dates, IDs) in a persistent block rather than losing them to summarisation.
• Escalation: honouring a request for a human, escalating policy gaps, and why sentiment and self-reported confidence are poor triggers.
• Error propagation across agents, and telling access failures apart from valid empty results.
• Managing context in large codebase exploration with subagents, scratchpad files and /compact.
• Human review workflows with calibrated confidence and stratified sampling.
• Preserving source attribution through multi-agent synthesis.

The six Foundations scenarios

Every Foundations question sits inside a scenario, and Anthropic publishes all six. You get four of them at random:

1. Customer support resolution agent
2. Code generation with Claude Code
3. Multi-agent research system
4. Developer productivity with Claude
5. Claude Code for continuous integration
6. Structured data extraction

Because you do not know which four you will get, you need to prepare for all six.

Officially out of scope for Foundations: fine-tuning, API authentication and billing, hosting MCP servers, embeddings and vector databases, computer use, vision, streaming, rate limits, cloud provider configuration (AWS, GCP, Azure), and prompt caching detail beyond knowing it exists.

Architect – Professional domains (CCAR-P)

Professional has seven domains, and Integration is the largest at 19%. These are the official weightings from Anthropic's exam guide:

Anthropic says it is for mid- to senior-level solution architects, AI/ML engineers, tech leads and senior software engineers. The guide explicitly says it is not for entry-level developers or people without experience designing end-to-end AI systems.

Domain 1: Solution Design & Architecture (17%)

This domain tests whether you can turn a business problem into the right Claude architecture. Tasks include designing end-to-end flows with feedback loops, choosing between workflow, agentic and augmented-LLM patterns, designing multi-agent orchestration, and tying the design to business value such as cost and performance SLAs.

Domain 2: Claude Models, Prompting & Context Engineering (13%)

This domain tests model selection and prompt design as trade-offs, not tricks. Tasks include choosing a Claude model for the job, designing system prompts and guardrails, zero-shot, few-shot and chain-of-thought prompting, managing context and token usage, and reuse through caching, modular prompts and Skills.

Domain 3: Integration (19%)

This domain tests whether you can connect Claude to enterprise systems safely. Tasks include:

• Spotting capability bloat in tool and agent configuration
• Finding authentication and authorisation gaps
• Justifying accuracy versus latency decisions
• Monitoring at scale
• Designing a RAG pipeline with suitable chunking and indexing, and matching retrieval to the data
• Choosing between MCP, API/CLI and agent-to-agent integration
• Choosing between progressive discovery and loading everything into context

Domain 4: Evaluation, Testing & Optimization (16%)

This domain tests whether you can prove a system works and find out why when it does not. Tasks include defining metrics across accuracy, latency, cost, safety and security, building evaluation datasets, A/B testing, diagnosing hallucinations and prompt failures, and optimising cost against performance.

Domain 5: Governance, Safety & Risk Management (14%)

This domain tests whether you can make a Claude system acceptable to a regulated business. Tasks include guardrails, identifying LLM failure modes, human-in-the-loop validation, compliance with regulations such as GDPR, HIPAA and FedRAMP, and bias, fairness and transparency.

Domain 6: Stakeholder Communication & Lifecycle Management (14%)

This domain tests the consulting half of the architect job. Tasks include structured discovery, explaining decisions and trade-offs, aligning expectations including SLAs, documenting architectures, and supporting the lifecycle from discovery through handoff, monitoring and iteration.

Domain 7: Developer Productivity & Operational Enablement (7%)

The smallest domain covers setting teams up to work with Claude. Tasks include configuring tools such as Claude Code for teams, improving developer workflows, and supporting debugging and operations.

How Foundations and Professional differ

Foundations asks "can you build it correctly?" Professional asks "should we build it this way, and can you defend that to the business?"

Two practical points follow:

1. Foundations rewards hands-on time with the tools. Many questions turn on exact behaviour, such as where a shared slash command lives or which flag stops Claude Code hanging in CI. You cannot reason your way to those; you have to have used them.
2. Professional rewards judgement. Its official sample questions are about least privilege, prompt caching to cut cost and latency, and diagnosing a RAG system after a document refresh. If you have never owned a production system end to end, Professional is the harder fit, whatever your job title.

There is no required order. If you are unsure, Foundations is the cheaper mistake.

How to pass: a domain-first prep plan

Prepare against the official task statements, weight your time by domain, and book only when your weakest domain is strong.

1. Download the official exam guide for your exam from the Anthropic Partner Academy and read it fully. The Foundations guide alone lists 30 task statements, with the knowledge and skills under each.
2. Take a diagnostic before any course. Answer questions across every domain so you know where you are weak.
3. Study the heaviest weak domain first. On Foundations, Agentic Architecture at 27% is the biggest lever. On Professional, Integration at 19% and Evaluation at 16% are.
4. Build, do not just read.
   • Foundations: follow Anthropic's own preparation exercises. Build an Agent SDK agent with escalation logic and a blocking hook. Configure Claude Code for a team with project CLAUDE.md, path-scoped rules, a forked skill and an MCP server. Build a structured extraction pipeline with a validation and retry loop. Build a coordinator with parallel subagents.
   • Professional: build and run one end-to-end Claude solution with RAG, an evaluation set and observability, as the official guide recommends. Then write down the architecture decisions and the trade-offs, as if explaining them to a client.
5. Do the official sample questions. Foundations has 12 with explanations, Professional has 3. Study why the wrong answers are wrong. The pattern repeats: the right answer usually fixes the root cause in the simplest proportionate way.
6. Turn every wrong answer into a flashcard and review it until it sticks.
7. Book on data, not on feel. My personal rule: track practice results by domain and book when every domain is consistently around 80%. That is my own rule, not an Anthropic standard, but it replaces guessing with evidence. With full-price retakes and waits of up to 90 days, guessing is expensive.

After you pass: recertification

Both credentials last 12 months, and on-time renewal is free.

You review what has changed since you certified and complete a free, non-proctored assessment on the Anthropic Partner Academy. If you let it lapse, you retake the full exam at full price. Anthropic also says that if the exam content changes significantly, it may require holders to retake the full exam instead.

Put the renewal date in your calendar the day you pass.

Preparing for Claude Architect with StudyTech AI

StudyTech AI has prep paths for both Claude Certified Architect – Foundations and Architect – Professional.

StudyTech AI is a personalised AWS and AI certification prep platform. It finds your knowledge gaps, builds your practice around your skills and gaps, turns your mistakes into flashcards, scores your readiness for every exam domain, and tells you when you are ready to take your exam.

For the Architect exams, that means:

• A free assessment by domain, mapped to the official Foundations or Professional domains
• Focused lessons on your weakest domain first, instead of studying everything equally
• Wrong answers become flashcards automatically
• Readiness by domain, the same breakdown your real score report uses
• Personalised mock exams built around your gaps

Pricing is simple: $29 a month or $297 a year.

StudyTech AI does not register you for the exam. You still need a Claude Partner Network company email to book through Pearson VUE.

Start a free Claude Architect assessment on StudyTech AI

Verified 7 November 2026 against Anthropic's official Architect – Foundations and Architect – Professional exam guides (version 1.0, effective July 2026), the Anthropic Partner Academy certifications page and FAQ, and Anthropic's Claude Partner Network announcement. Both guides say they are subject to change without notice, so check the current version on the Anthropic Partner Academy before you book.