What is the Claude Certified Architect certification?

The Claude Certified Architect, Foundations (CCAR-F) is Anthropic's first technical certification, launched March 12, 2026. It is a 60-question, 120-minute proctored multiple-choice exam testing production knowledge of the Claude Agent SDK, Claude Code, the Claude API, and MCP across five weighted domains. It costs $125, requires a 720 scaled score to pass, and the credential is valid for 12 months.

Who is eligible to take the Claude certification exam?

Certification is currently open to organizations in the Claude Partner Network, which is free to join for any company bringing Claude to market. Candidates register through the Anthropic Partner Academy and take the exam online via Pearson OnVUE or at a test center. Anthropic has announced the program spans three roles, Practitioner, Architect, and Developer, with more certifications rolling out later in 2026.

Is the Claude Certified Architect exam hard?

It targets solution architects with six or more months of hands-on production experience with Claude. Questions are scenario-based with distractors designed to catch tutorial-level knowledge, and the pass mark is 720 on a 100 to 1,000 scale. Retakes are allowed after 14, 30, then 90 days, capped at four attempts per rolling year.

Does the CCA certification cover security and compliance?

No. The five domains cover architecture, tooling, Claude Code, prompting, and reliability. Securing the code that gets built, governing the data MCP connections touch, and producing audit evidence are outside its scope, which is exactly the layer enterprises still need to add around certified builders.

Claude Certified Architect: What the CCA Exam Covers and Who Should Take It

The short answer

CCAR-F is a 60-question, 120-minute proctored exam, $125, five domains weighted toward agentic architecture (27%). Open to Claude Partner Network members (free to join). Worth it for architects with 6+ months of hands-on Claude production work; premature for anyone still in tutorials.

On March 12, 2026, Anthropic launched its first technical certification, the Claude Certified Architect, Foundations (exam code CCAR-F), alongside the Claude Partner Network and a committed $100 million for partner training, technical support, and market development in 2026. Search interest spiked to roughly 24,000 queries the month it launched and has stayed in five figures since. This guide covers what the exam actually tests, who should take it, and where a certificate genuinely helps, based on Anthropic's official exam guide rather than secondhand summaries.

The exam at a glance, from the official guide

Format: 60 multiple-choice questions in 120 minutes, each with one correct answer and three plausible distractors. The exam presents 4 scenarios drawn at random from a bank of 6, and every question is framed inside one of those production contexts.

Delivery and cost: online proctored through Pearson OnVUE or at a test center. The fee is $125 USD. Worth noting: several popular blog posts list the fee as $99, which does not match Anthropic's current exam guide. Trust the guide.

Scoring: scaled 100 to 1,000, with 720 to pass, reported as pass or fail. The credential is valid for 12 months.

Retakes: 14 days after a first failed attempt, 30 after the second, 90 after the third, with a maximum of four attempts in any rolling 12 months.

Eligibility: certification is currently open to organizations in the Claude Partner Network, which is free to join for any company bringing Claude to market. Badges are issued digitally through Credly, and preparation courses live in the Anthropic Partner Academy.

The five domains, with weightings

The exam covers five content domains. The weightings tell you where to spend preparation time:

1. Agentic Architecture and Orchestration, 27%. The largest domain: multi-agent design with the Claude Agent SDK, subagent delegation, tool integration, lifecycle hooks, and when a single agent beats a swarm.

2. Claude Code Configuration and Workflows, 20%. CLAUDE.md files, Agent Skills, custom slash commands, plan mode versus direct execution, and wiring Claude Code into team workflows. If your team already runs Claude Code at enterprise scale, this domain will feel familiar.

3. Prompt Engineering and Structured Output, 20%. JSON schemas, few-shot patterns, extraction from unstructured documents, and prompts that survive production rather than demos.

4. Tool Design and MCP Integration, 18%. Designing Model Context Protocol tools and resource interfaces for backend systems. This is the domain closest to real enterprise value, and the one where exam knowledge and production reality diverge most, more on that below.

5. Context Management and Reliability, 15%. Long documents, multi-turn conversations, multi-agent handoffs, error handling, and human-in-the-loop escalation.

The six scenarios

Every question lives inside a realistic production scenario: a customer support resolution agent with MCP tools like process_refund and escalate_to_human, Claude Code in daily development, a multi-agent research system with coordinator and specialist subagents, developer productivity tooling over legacy codebases, Claude Code inside CI/CD for automated review and test generation, and structured data extraction with JSON validation. You get four of these six at random, so preparing means being comfortable in all of them.

Who should take it, and who should not

Anthropic's target candidate is specific: a solution architect with six or more months of hands-on experience building production applications with the Claude API, Agent SDK, Claude Code, and MCP. If that describes you or your team, the exam is a reasonable validation and, inside the Partner Network, counts toward partner standing.

Skip it, at least for now, if you are earlier in the journey. The exam assumes production judgment, not tutorial knowledge, and the 12-month validity means certifying before you build is mostly paying $125 twice. Anthropic has said additional certifications for sellers, architects, and developers are coming later in 2026, and the program lists three roles overall: Practitioner, Architect, and Developer. Early Partner Network members get priority access as those roll out.

A disclosure, since we have skin in this game: every engineer Clarista staffs on client engagements holds this credential. It is part of how we staff AI development work for regulated firms, and it shaped the opinions in this guide.

What the certificate does not cover

Read the domains again and notice what is absent: nothing on securing what gets built, and nothing on governing the data these systems touch. That is not a flaw, it is scope. The CCA validates that an architect can design a Claude application; it says nothing about whether the code that ships is scanned for vulnerabilities, whether the MCP connection to your warehouse masks PII, or whether an auditor gets evidence.

For enterprises, those absences are exactly where projects stall. A certified architect can build you a beautiful agent; the security review still asks who can access what, and the compliance team still wants the audit trail. That layer is what we build at Clarista: scanning for everything AI writes, a governed MCP connection for everything AI reads, and compliance evidence for both. Certification and governance are complements: one proves the builder, the other protects the build.

How to prepare, practically

The official exam guide (37 pages, free) is the primary source, with domains, task statements, and sample questions. The Anthropic Partner Academy carries self-paced preparation courses for network members. But the guide itself is blunt that hands-on experience is the real preparation: the distractors are designed to catch people who have read about Claude but not shipped with it. Six months of building, especially anything involving MCP tools and multi-agent orchestration, beats any amount of cramming.

If you would rather ship first and certify later, that instinct is sound. Teams that want production Claude applications now, with the security and compliance layer already handled, can also simply hire AI developers who build on a governed platform and hand over audit-ready work.