Claude Certified Developer – Foundations: what the exam tests

The Claude Certified Developer – Foundations exam (CCDV-F) tests whether you can build and ship an application on Claude, and most of its marks are on ordinary engineering. Applications and integration is 33.1% of the 53 questions: API mechanics, application design, configuration and plain software engineering. Claude Code, which many people expect to dominate, is 3.1%.

The exam runs 120 minutes and costs $125. The pass is a scaled 720 out of 1,000, and each question says how many answers to choose. The guide suggests 1 to 5 years of software work and 6 months with Claude or a similar LLM system. As with every Claude certification, only staff of Claude Partner Network organisations can register; our comparison of the 4 exams covers that.

Where the marks are

The guide splits its 8 domains into 25 skills, each with its own weight. Switch the view to see them ranked.

Try itWhere the 53 questions go

Show

·domains·skills

Applications and Integration33.1% ≈17.5

Model Selection and Optimization16.8% ≈8.9

Agents and Workflows14.7% ≈7.8

Prompt and Context Engineering11% ≈5.8

Tools and MCPs10.6% ≈5.6

Security and Safety8.1% ≈4.3

Claude Code3.1% ≈1.6

Eval, Testing, and Debugging2.6% ≈1.4

Two things stand out. Software engineering foundations (REST, JSON, async code, version control, code review, refactoring) is 7.4%, and technical fundamentals is another 6.1%. That is about 7 questions on general engineering you cannot learn from Claude's docs. And the domains people prepare hardest for are small: Claude Code 3.1%, debugging 2.6%, hooks 1%.

What the sample questions reward

The guide ships 3 sample questions, and all 3 have the same shape. The wrong answers adjust the model or the prompt. The right one changes the system around it.

• An overnight job of 10,000 documents where cost comes first. The answer is the Message Batches API, at half the standard price with results within 24 hours. Running calls in parallel, lowering max_tokens or picking the smallest model are the distractors.
• A page with hidden text telling the agent to reveal its system prompt. The answer is to treat the page as untrusted input, keep it apart from your instructions, and use guardrails or hooks so it cannot trigger anything sensitive. The guide adds that a model that follows instructions better can be more open to injection, not less.
• An internal REST service several apps need. The answer is an MCP server that each app connects to, rather than code in each app's prompt.

Expect the rest of the exam to work the same way. When 3 answers tune the model and 1 adds a control, the control is usually right.

Where the guide's wording is behind the docs

The guide is version 1.0 from July 2026. A few of its words do not match Anthropic's documentation today. Learn the guide's answer for the exam and the docs' for work.

• Websockets. The guide lists websockets under technical fundamentals. The Messages API streams over server-sent events, not websockets.
• Sockets for MCP. The guide names stdio and sockets. The current MCP specification's transports are stdio and Streamable HTTP.
• Opus, Sonnet and Haiku. The guide compares the 3 families. Anthropic's model guide now adds Claude Fable 5.1 above Opus 5.5 for the hardest reasoning and long agent runs, and suggests trying a higher effort setting before a bigger model.
• Commands. The guide lists commands as a Claude Code component. Commands have since been merged into skills, though .claude/commands/ files still work.

How to prepare

Build one application end to end on the API: streaming, tool use, structured output, prompt caching and a batch job, with an MCP server it calls. In our AI engineering course, the LLM API module covers the mechanics and tools and agents the agent half. Spend a week on prompt caching and token costs, since cost and model choice together come to 5.5%. If your engineering basics are rusty, the 13.5% of general engineering is the cheapest place to gain marks.

Then try the free Developer – Foundations practice questions, written to the guide's skills. If you mostly design agent systems rather than ship applications, the Architect – Foundations guide fits better.