Anthropic now runs a certification program for Claude, and one of its four exams is written for engineers: Claude Certified Developer – Foundations, exam code CCDV-F. Many vendor exam guides describe their content in round numbers and loose themes. This one publishes a weight for every skill, down to a tenth of a percent, which makes it possible to see exactly where the points are before you study anything.
This article reads the official exam guide (version 1.0, effective July 2026) and the program FAQ the way you would read a spec: what is in scope, how much each part is worth, and what the published sample questions reveal about how you will be tested. It also covers the three sibling exams, because the four are aimed at very different people.
First, check that you are allowed to sit it
The program is not open to the public yet. Anthropic's FAQ states that certification is currently available only to organizations in the Claude Partner Network, and that registration requires a partner email address on a recognized company domain. Personal email addresses do not work.
If your employer is not a partner, you cannot register today. The exam guide is still worth reading, since it is a detailed public statement of what Anthropic expects a competent Claude developer to know.
The four exams at a glance
| Exam | Code | Written for | Questions | Fee |
|---|---|---|---|---|
| Associate – Foundations | CCAO-F | Professionals who use Claude as a productivity tool | 60 | $99 |
| Developer – Foundations | CCDV-F | Engineers who build with the Claude API, Claude Code, and MCP | 53 | $125 |
| Architect – Foundations | CCAR-F | Solution architects who design and implement production applications | 60 | $125 |
| Architect – Professional | CCAR-P | Mid- to senior-level technical professionals who design and deliver production systems | 63 | $175 |
Everything else is shared. Each exam runs 120 minutes, uses multiple-choice and multiple-response questions, and is passed with a scaled score of 720 on a scale of 100 to 1,000. Pearson VUE delivers all four, online with a proctor or at a test center, in English or Japanese. A credential is valid for 12 months.
The fees are list prices. Partners on the Registered tier pay them in full, while the Select, Preferred, and Global Premier tiers get 50% off at checkout. Even at full price, $99 to $175 is toward the lower end of what IT certification exams usually cost.
The Developer blueprint, domain by domain
| Domain | Weight |
|---|---|
| Applications and Integration | 33.1% |
| Model Selection and Optimization | 16.8% |
| Agents and Workflows | 14.7% |
| Prompt and Context Engineering | 11.0% |
| Tools and MCPs | 10.6% |
| Security and Safety | 8.1% |
| Claude Code | 3.1% |
| Eval, Testing, and Debugging | 2.6% |
The guide says these percentages approximate the share of scored questions drawn from each domain. With 53 questions on the exam, that turns the table into a rough question count, and a few things stand out.
Half the exam is two domains. Applications and Integration plus Model Selection and Optimization come to 49.9%. If you are short on time, this is where it goes.
Claude Code and evals are almost absent. At 3.1% and 2.6%, each works out to one or two questions. If you came to Claude through Claude Code, that may be the opposite of what you expected. The exam is about building applications on the platform, and the coding tool is a small part of that.
About a fifth of the exam is not about Claude at all. Four skills cover general engineering: Software Engineering Foundations (7.4%), Technical Fundamentals (6.1%), Understanding Requirements (3.4%), and Systems Life Cycle (2.8%). Together they make 19.7%. The guide lists REST APIs, JSON, asynchronous programming, version control, code review, and refactoring under these headings. An experienced engineer already has those points. Someone who learned prompt writing without a software background does not.
The largest single skills are design and mechanics. Claude Application Design is worth 8.6%, and Claude API Mechanics is worth 6.8%. The first covers how Claude interprets instructions across interfaces, content boundaries, schema design, and session hygiene. The second covers messages, tools, streaming, vision, thinking, caching, and the trade-off between the real-time and batch APIs.
Choosing a tool type outweighs writing one. In the Tools and MCPs domain, MCP Server Development is 2.1%. Agentic Customization, which is the judgment call between built-in tools, custom tools, Skills, and MCP servers for a given use case, is 4.1%. The exam cares more about whether you pick the right mechanism than whether you can author every kind.
What the questions look like
The guide includes three sample questions and notes that they are illustrative and not taken from the live question bank. They share a pattern that is worth knowing.
One describes a developer who must process 10,000 documents overnight for a non-urgent report, with cost as the main concern. The correct answer is the Message Batches API. The wrong answers are things people really do: firing every request synchronously in parallel, lowering max_tokens, or dropping to the smallest model regardless of output quality.
Another describes an agent that summarizes user-submitted web pages, one of which hides an instruction to reveal the system prompt. The correct answer treats page content as untrusted input, keeps it apart from trusted instructions, and uses guardrails or hooks so injected text cannot trigger sensitive actions. The distractors are raising the temperature, adding a polite line to the system prompt, and switching to a larger model.
In both cases every wrong option sounds like a fix and addresses a different problem. The questions test whether you can match a mechanism to a requirement, so memorized definitions will not carry you far.
Who the exam assumes you are
There are no mandatory prerequisites. The guide does describe the candidate it was written for:
- One to five years of software engineering experience
- At least six months of hands-on work with Claude or a comparable LLM system
- Proficiency in Python, TypeScript, or both
- Fluency with REST APIs and command-line tools
- A working understanding of LLM fundamentals, agents, context management, and MCP
It also says plainly who the exam is not for: non-technical or casual users, and roles limited to prompt writing without broader application development responsibility.
How to prepare
Anthropic states that no single course is required and that no resource guarantees a pass. The guide's own preparation list is short and practical. Its central item is to build and operate at least one Claude application that exercises the API, integrates one or more tools, applies basic prompt and context engineering, and includes simple security and evaluation practices.
A sensible order, following the weights:
- Start with the API: messages, tool use, streaming, caching, and when batch beats real time.
- Add model selection and cost. Know the use cases for Opus, Sonnet, and Haiku, and how token budgeting and prompt caching change a bill.
- Build one agent with a tool-use loop, then decide where a plain workflow would have been enough.
- Write one custom tool and connect one MCP server, so the trade-offs between them are concrete.
- Read the security skills last but carefully: prompt injection, untrusted input, least privilege, and secrets handling together make up 8.1%.
There is no official practice exam at the moment. The FAQ says the earlier one was retired when delivery moved to Pearson, and it points candidates to the sample questions in the guide.
Rules that cost money
- Retakes are paid in full. Waiting periods are 14 days after a first failed attempt, 30 after a second, and 90 after a third, with at most four attempts in a rolling 12 months.
- The reschedule deadline depends on where you test. The exam guide says 24 hours. The FAQ is more specific: 48 hours before a test center appointment and one hour before an online one. Miss the window or fail to show up, and the fee is forfeited.
- The exam is closed book. Notes, documentation, translation tools, and AI assistants are all prohibited.
- Renewal is cheaper than lapsing. On-time renewal is a free, non-proctored assessment. If the credential lapses, you retake the full exam at the full fee.
Which exam should you take?
If you write code against the API, take Developer – Foundations. If your work is designing systems end to end, look at the two Architect exams. Architect – Foundations is built around scenarios, with four drawn from a bank of six, and its largest domain is Agentic Architecture and Orchestration at 27%.
The Associate exam is the one to recommend to non-technical colleagues. Its guide says it is not intended for software developers, and it is the only one of the four that does not count toward a firm's partner tier. For that exam, EveryExamPrep has a free Claude Certified Associate guide with 150 practice questions written from the official blueprint.
The short version
Developer – Foundations is 53 questions in 120 minutes for $125, open only to Claude Partner Network organizations. Half of it is application integration and model selection, a fifth is general software engineering, and Claude Code barely appears. Build one real application before you book, and study in the order the weights suggest.
