Anthropic certification preparation guide
How the Claude certifications work, which one fits you, what the Architect exam actually tests, and a six-week plan to pass on your first attempt.
Updated · ClaudeMock team
In March 2026 Anthropic launched its first technical credential, the Claude Certified Architect — Foundations. It doesn’t ask whether you can write a clever prompt. It asks whether you can design a Claude system that survives production: agents that stop when they should, tools that fail safely, and context that doesn’t quietly fall apart halfway through a long task.
This guide walks through every Claude certification, explains the Architect blueprint domain by domain, and gives you a study plan built around the way the exam is written.
Choose your certification
The certifications build on each other. Pick the one that matches the work you do today, not the job title you want next.
Claude Certified Associate — Foundations
For people starting out with Claude who want to prove they can use it well in everyday work: prompting, choosing the right model and product, and working with AI responsibly.
Claude Certified Developer — Foundations
For engineers who build with the Claude API: messages, tool use, structured output, streaming and error handling in real applications.
Claude Certified Architect — Foundations
For solution architects designing production systems with Claude Code, the Claude Agent SDK, the Claude API and MCP. Anthropic’s first technical certification.
Claude Certified Architect — Professional
The advanced architect track for people who already design Claude systems and want to show depth in multi-agent design, governance and scale.
Exam format
Here is what candidates report for the Architect — Foundations exam. Anthropic updates its exams, so confirm the details in the official exam guide when you register.
| Questions | 60 scenario-based multiple-choice questions |
|---|---|
| Time | 120 minutes, about two minutes per question |
| Passing score | 720 on a scale of 1,000 |
| Delivery | Online, proctored |
| Scenarios | Questions are grouped around realistic production scenarios, drawn at random from a larger pool |
The one thing to know before you study
The exam is about judgment, not recall. Several answers will work in a demo; only one holds up in production. Expect many questions where the right move is to add a deterministic check, hand off to a human, or not use an agent at all.
What each domain tests
Agentic architecture & orchestration
The heaviest domain. Know when a fixed workflow beats an autonomous agent, how the agent loop decides to continue or stop, and how to split work between an orchestrator and subagents. Be ready to spot runaway loops, missing stop conditions and places where a human must approve an action.
Claude Code configuration & workflows
Project memory files, permission settings, hooks, custom slash commands, subagents and running Claude Code headless in CI. Questions often ask where a rule belongs: in memory, in a hook, or in a permission setting.
Prompt engineering & structured output
Clear system prompts, examples, XML tags, and getting reliable JSON through tool schemas or structured output. Learn how to validate output and what to do when it fails validation.
Tool design & MCP integration
Writing tool names and descriptions Claude can use correctly, designing input schemas, returning useful errors, and choosing between tools, resources and prompts in an MCP server. Security matters here: least privilege and treating tool output as untrusted data.
Context management & reliability
Keeping long tasks on track: summarising and compacting context, prompt caching, retrieval, retries and fallbacks, and logging enough to debug an agent after the fact.
Six-week study plan
Plan for five to eight hours a week. Each week ends with something you build, because building is what the exam rewards.
-
1
Baseline and API fundamentals
Take a full practice exam cold to find your weak domains. Then work through the Messages API, model choice and system prompts. Build a small script that calls Claude and returns validated JSON.
-
2
Tool use and MCP
Add two tools to your script, then move them into a small MCP server. Break a tool on purpose and make sure Claude gets an error it can act on.
-
3
Agents and orchestration
Build an agent loop with the Claude Agent SDK. Add a stop condition, a step budget and a human approval step for anything destructive. Try an orchestrator with two subagents.
-
4
Claude Code in depth
Set up a repository with a project memory file, permission rules, a pre-tool hook and a custom command. Run Claude Code non-interactively as part of a CI job.
-
5
Context and reliability
Give your agent a task long enough to strain its context. Add summarisation, prompt caching, retries and logging, and compare cost and accuracy before and after.
-
6
Timed practice and review
Take two or three full timed practice exams. For every miss, write one sentence on why the right answer wins in production. Book the exam once you score comfortably above the pass mark.
Study resources
Start with Anthropic’s own material: the official exam guide, the Claude documentation for the API, Claude Code and MCP, and the free courses on Anthropic’s learning platform, such as Claude 101, AI Fluency, building with the Claude API and Claude Code in Action.
Then test yourself. Our study materials summarise each domain, and the study guide maps topics to the official blueprint. When you’re ready, sit a timed mock exam.
Exam day
Run the proctoring system check the day before, on the same computer and network you’ll use. Clear your desk and close every other application.
Read the scenario before the question. Most wrong answers are reasonable ideas that ignore one constraint in the scenario, such as a cost limit, a compliance rule or an unreliable tool.
Keep moving. At two minutes a question, flag anything that takes longer and come back to it. Between two plausible answers, prefer the one that is safer, easier to observe and simpler to operate.
Find out where you stand today
Take a timed practice exam built on the official domains, with explanations for every answer.
Frequently asked questions
Which Anthropic certification should I take first?
If you use Claude day to day but don’t write code, start with the Associate exam. If you build on the API, take Developer. If you design systems that other engineers build, go straight to Architect — Foundations; Architect — Professional makes most sense once you hold it.
How long does it take to prepare?
Most working engineers need four to six weeks of steady study for the Architect — Foundations exam, less if they already ship Claude-based agents at work. The six-week plan above is a good default you can compress.
Is hands-on experience really necessary?
Yes. Questions describe a production situation and ask for the best architectural decision, so you need to have seen agents loop, tools fail and context overflow. Building one small agent end to end teaches more than rereading documentation.
Are practice exams worth it?
They are the best way to find weak domains and get used to the time pressure. Treat each wrong answer as a study prompt: work out why the right option is right and why the others fail in production.
What happens if I fail?
You can retake the exam after a waiting period that gets longer with each attempt. Check Anthropic’s current exam policy for exact retake windows and credential validity before you book.
ClaudeMock is an independent practice-exam provider and is not affiliated with or endorsed by Anthropic. Claude and Anthropic are trademarks of Anthropic PBC.