I passed the Claude Certified Developer—Foundations exam with 896 out of 1000, and I wrote down how I got there so you don’t have to piece it together from five different pages. Friend’s link
I’m not affiliated with Anthropic, and I won’t share any exam questions. I checked the exam details below against the official certification page and FAQ.
What is the exam about?
The exam is for engineers who build with the Claude API, Claude Code, and the Model Context Protocol (MCP), from a first integration all the way to agents running in production. Writing good prompts is only a small part of it. Most of the questions describe a situation and ask which approach fits and why.
Important eligibility note: As of 26 October 2026, Anthropic’s certification program is restricted to people at Claude Partner Network organizations. Registration requires a recognized company-domain email address. Check the current eligibility rules before you buy the exam. Here are the practical details from the official FAQ:
Format: The FAQ lists multiple-choice and scenario-based multiple-response questions. My attempt had 53 questions, all multiple-choice
Time: 120 minutes to answer. Plan for about 135 minutes in total, because check-in and instructions take time.
Where: online with OnVUE proctoring or at a Pearson test center. It’s closed book.
Cost: $125.
Who can take it: you need to be 18 or older and carry a government-issued ID. Read the eligibility note above and the “who can take an exam” page before you plan anything else.
Validity: 12 months. Renewal is a free, non-proctored assessment.
The official source is the Anthropic certification page, and the certification FAQ has the rules.
Syllabus and topics, at a high level
For the official domains and how much each one counts, open the exam guide PDF linked from the certifications page. This is the list my own notes were built around:
Tokens, context windows, sampling, and model selection
Prompt engineering, XML tags, and structured outputs
SDKs, REST APIs, streaming, and asynchronous processing
Tool schemas and the tool-use lifecycle
Agents, workflows, orchestration, and human approval
Persistent memory, context compaction, and skills
Images, PDFs, and batch processing
Security, evaluation, testing, and debugging
Trade-offs show up everywhere in these topics. Four kept coming back in my notes:
Workflow or agent? When the steps never change, a plain workflow is usually the better design. An agent makes sense when the next step depends on what just happened.
Sync, streaming, or batch? Ask whether a user is waiting, whether partial output helps, and whether the work can finish later.
Long conversations. Each turn makes you decide what to keep, summarize, store, or drop, and every choice costs something.
Tool use. The model requests a tool call, but your application runs it and returns the result. Clear tool descriptions, error handling, stop conditions, and human approval for risky actions all matter.
Passing marks
You get a scaled score from 100 to 1,000, and the minimum passing score is 720. "Scaled" means it isn’t the percentage of questions you got right. The FAQ doesn’t list a question count, but my exam had 53 multiple-choice questions.
I scored 896, which is 176 points over the line. I wouldn’t aim for 720 and hope, though. A failed attempt means a wait: 14 days after the first, 30 after the second, 90 after the third, and no more than four attempts in any 12 months. Each retake costs the full fee.
My preparation strategy: one topic, one set of notes, then questions
The best thing I did was stop treating the syllabus as one huge subject. I split it into sections and added each one to the same master document, so the notes grew as I understood more. Every section I added followed the same layout:
Core concepts and the technical distinctions between ideas that look alike
Memory hooks for the things that are easy to forget
Common mistakes and exam traps
Active-recall questions
Scenario-based multiple-choice questions, each with an explanation of why the answer is right
A quick revision card for the last few days before the exam
You can copy that layout today and use it for any syllabus. I also answered five questions for every section:
What does this mean?
Why does it matter?
When would I use it?
What can go wrong?
How would I spot the right approach in a real scenario?
The last one changed how I revised. Memorizing a definition got me nowhere, so I had to picture the concept running inside a real application.
Every section also got its own practice questions. For active recall, I explained the concept without looking at my notes. For scenarios, I read a situation, picked an approach, and wrote down why the other options were weaker.
Want to study with my notes?
Get my Claude Developer Foundations Study Notes →
These are independently prepared personal study notes, not official Anthropic material. They are intended to support your learning and revision; they do not contain a promise of exam success or guarantee a passing score.
Keep an error log
If I started again, I’d keep an error log from day one. Whenever you get a practice question wrong or get it right for the wrong reason, write down three things:
Concept: what the question was really testing
My wrong assumption: what I believed at the time
Rule to remember: the decision rule that would have given me the right answer.
Reading that log is a better use of an evening than rereading a section you already know.
The 5-step study plan I would follow again
This is the short version of everything above, in the order I’d do it again.
Complete the official prep course. Start with the official developer prep course. It’s the closest thing to the real scope of the exam, so don’t swap it for a video or a cheat sheet.
Take some mock tests. Practice exams show you the gaps your notes hide. Look for questions written specifically for the Developer—Foundations exam. Use every mock question to find a weak spot, and don’t expect the real exam to look the same.
Use Claude to practice whatever you learn. Take a topic you just studied and ask Claude to quiz you on it. Ask for scenario questions, answer them yourself first, and only then read the explanation. Check anything version-specific against the current documentation, because practice questions from any AI are study aids and not official exam questions.
Register for the exam. Once the course, the mock tests, and your notes feel solid, register here. It costs $125, so check eligibility first, then choose online proctoring or a test center.
Be calm. By now you’ve done the course, the mock tests, and the notes. The exam is closed book, and you have 120 minutes. Get your ID ready, and if you’re testing online, check your setup the day before. Read each scenario all the way through before you look at the options.
Top comments (0)