Claude Certified Architect certification: the complete guide
Everything you need to know about Anthropic's Claude Certified Architect – Foundations exam: what it tests, the five domains and their weights, how the exam works, and exactly how I prepared and passed with 833 out of 1000. Written by someone who sat it, not summarized from a press page.
The short version: the Claude Certified Architect – Foundations exam (code CCAR-F) is Anthropic's certification for people who design and ship agentic systems on Claude. It's 60 multiple-choice questions in 120 minutes, proctored online, scored 100 to 1000, and you pass from 720. I took it on July 15, 2026 and scored 833. This guide walks through what it actually tests, who it's for, how to prepare, and whether it's worth your time.
What the Claude Certified Architect certification is
Anthropic launched the certification on March 12, 2026. It isn't a trivia quiz about the Claude API. It's an architecture exam: it checks whether you can make the right design decisions when you build agents, tools and workflows that have to survive production. Almost every question hands you a realistic scenario and several technically valid options, then asks which one fits the constraints. That framing — decisions, not definitions — is the single most important thing to understand before you sit it. I unpack what that felt like in my write-up of passing the exam.
"Foundations" is the entry tier. It establishes that you understand the building blocks of agentic engineering on Claude well enough to make sound calls under real constraints. It's aimed at engineers, architects and technical leads who work with Claude, the Claude Agent SDK, MCP and Claude Code — not at people who have only read the marketing.
The five exam domains
The exam guide splits the material into five domains, each with a fixed weight. Those weights tell you where to spend your study time — over a quarter of the exam is a single domain. Here is the official blueprint:
| Domain | Weight | What it covers |
|---|---|---|
| D1 · Agentic Architecture & Orchestration | 27% | Driving the agentic loop on stop_reason, multi-agent design, human handoff, guardrails and bounded autonomy. |
| D2 · Tool Design & MCP Integration | 18% | Designing tools an agent can use reliably, MCP servers, tool_choice, and error surfaces that don't fail silently. |
| D3 · Claude Code Configuration & Workflows | 20% | CLAUDE.md, hooks, subagents, permissions and plan mode — configuring Claude Code so work is safe and repeatable. |
| D4 · Prompt Engineering & Structured Output | 20% | Reliable prompting, extended thinking, JSON/structured output, and stopping the model from fabricating missing data. |
| D5 · Context Management & Reliability | 15% | Context windows, compaction, prompt caching, and keeping long-running sessions coherent and cost-controlled. |
If you want the condensed heuristics behind these domains — the recurring decisions that separate a right answer from a plausible one — I wrote them up as 12 rules for building agents. They came straight out of preparing for this exam and cover a large slice of D1 and D2.
Who should take it
This certification is a fit if you already build with Claude and want a credential that reflects real design skill rather than tool familiarity. Concretely, good candidates:
- ship agents, tools or workflows on Claude, the Agent SDK, MCP or Claude Code in actual projects;
- have hit production problems — silent tool failures, runaway loops, context blowups — and had to fix them;
- can reason about trade-offs (deterministic guardrail vs prompt instruction, compaction vs fresh session) rather than reciting definitions.
You're probably not ready yet if your Claude experience is a few chat sessions and a tutorial. The exam assumes hands-on scars. If that's you, the fastest way to close the gap is to build something small end-to-end first, then come back.
This is what I do hands-on — advising on AI strategy and building agents that survive the demo.
How the exam works
The format is fixed and worth knowing before you book:
Every question is single-best-answer: four options, exactly one correct, three plausible distractors. The scoring is scaled — 100 to 1000, not a raw percentage — and you pass from 720. It's proctored online, so you sit it in one uninterrupted 120-minute window with your webcam on. There's no partial credit and no penalty structure beyond a wrong answer scoring zero, so leaving a question blank is never better than guessing. Registration goes through Anthropic's certification program; check the official Anthropic Academy page for current availability, eligibility and pricing, because those details can change.
How to prepare
Here's the plan I'd give anyone starting now. It's the one I effectively followed to a 833.
PREP PLAN
- 1. Read the official exam guide first. Know the five domains and their weights before anything else. Over 45% of the exam is D1 and D3 combined — budget accordingly.
- 2. Build one thing end-to-end. An agent with a couple of MCP tools, a hook, a Claude Code config. Nothing teaches D1–D3 like hitting the failure modes yourself.
- 3. Drill decision-making, not vocabulary. Practice picking the right option among several valid ones. My free 60-question practice exam is built exactly for this.
- 4. Simulate the real thing once. A single 120-minute sitting, no breaks, exam mode. You're training stamina and pacing, not just knowledge.
- 5. Patch your two weakest domains. Use the per-domain breakdown to see where you leak points, then go back to the docs for those areas only.
The practice exam is the centerpiece of that plan. It runs in the browser, it's free, and it mirrors the real format — 60 original scenario questions, a study mode with explanations and an exam mode with the 120-minute clock, scored on the same 100–1000 scale with a per-domain breakdown.
Practice with 60 questions, free
Study and exam modes, scored 100–1000 with a per-domain breakdown. In the browser, no signup.
Is it worth it?
For me, yes — with a caveat. The value isn't the badge on its own; it's that preparing forces you to systematize knowledge you picked up in scattered pieces. Studying for CCAR-F made me articulate why a deterministic hook beats a polite prompt, when to compact versus fork a session, and how to design a tool that fails loud instead of silent. That clarity carries into the work.
As a signal, it's most useful if you sell or lead AI implementation work, where a client wants evidence you can architect systems that reach production, not just prototype them. It pairs naturally with the Claude partner credentials. If you're a hobbyist with no need to prove anything to anyone, the free practice exam alone will give you most of the learning without the fee.
Frequently asked questions
What is the Claude Certified Architect certification?
It's Anthropic's certification (Foundations tier, exam code CCAR-F) for engineers and architects who design agentic systems on Claude. It tests design decisions across five domains — agentic architecture, tool and MCP design, Claude Code configuration, prompt engineering and structured output, and context management — rather than API trivia.
How many questions are on the exam and how long is it?
Sixty single-best-answer multiple-choice questions in a 120-minute, proctored online session. Each has four options with exactly one correct answer.
What score do you need to pass?
The exam is scored on a scale from 100 to 1000, and the passing score is 720. There's no partial credit, so an unanswered question is never better than a guess.
What are the five exam domains and their weights?
Agentic Architecture & Orchestration 27%, Claude Code Configuration & Workflows 20%, Prompt Engineering & Structured Output 20%, Tool Design & MCP Integration 18%, and Context Management & Reliability 15%.
How long does it take to prepare?
It depends on your hands-on experience. If you already build with Claude, MCP and Claude Code, a focused week of drilling decision-style questions and simulating one full sitting is usually enough. If you're newer, build a small end-to-end agent first — the exam rewards scars, not reading.
Is there a free practice exam?
Yes. I wrote a free 60-question practice exam that runs in the browser with study and exam modes, scaled scoring and a per-domain breakdown. It's unofficial, modeled on the public exam guide, and needs no signup.
Is the certification worth it?
If you sell or lead AI implementation work, the credential is a useful signal that you can architect production systems, and preparing for it sharpens real skills. If you just want to learn, the free practice exam captures most of the value without the fee.
Disclaimer. This guide is independent and unofficial. It is not affiliated with, authorized or endorsed by Anthropic. Exam details are modeled on the publicly available exam guide and my own experience sitting it; check Anthropic's official channels for current specifics. Claude and Claude Certified Architect are names owned by Anthropic.
