Learn Claude Code

Claude Certified Developer – Foundations

Anthropic's certification for developers who build on Claude. A third of the exam is about building applications on the API, so it rewards hands-on API experience more than Claude Code knowledge.

Take the 53-question mock examPractise by domain
At a glance

53 questions in 120 minutes, scored out of 1000 with 720 to pass. Formats include multiple choice, select-TWO, Yes/No matching (judge each statement in a teammate's notes) and classification (sort items into categories). Applications and Integration is a third of the exam. Time limit, question count and pass mark are as widely reported for exam guide v1.0; confirm on Anthropic's certification pages before you book.

The domains and their weights

Agents and Workflows14.7%
Applications and Integration33.1%
Claude Code3.1%
Eval, Testing, and Debugging2.6%
Model Selection and Optimization16.8%
Prompt and Context Engineering11%
Security and Safety8.1%
Tools and MCPs10.6%

Agents and Workflows 14.7%

  • Use a fixed workflow when the steps are known; use an agent when the path depends on what it finds.
  • Run agent loops on stop_reason, not on parsing the text.
  • Hand noisy work to subagents so the main context stays focused.
  • Put a person in the loop for irreversible actions.

Practise 12 Agents and Workflows questions →

Applications and Integration 33.1%

  • Know the Messages API: roles, the system prompt, and sending the full history each turn.
  • Tool use: Claude returns tool_use blocks; your code runs the tool and replies with a tool_result carrying the matching tool_use_id.
  • Use tool schemas for reliable structured output.
  • Stream for responsive interfaces; retry rate-limit and overload errors with backoff.
  • Prompt caching for repeated long prefixes; Message Batches for large jobs that can wait.

Practise 26 Applications and Integration questions →

Claude Code 3.1%

  • CLAUDE.md for project conventions; permissions to control what Claude Code can do.
  • claude -p runs it non-interactively in scripts and CI.

Practise 3 Claude Code questions →

Eval, Testing, and Debugging 2.6%

  • Build a representative test set before changing prompts or models.
  • Re-run it after every change to catch regressions.
  • Debug by reading the full transcript, not only the final answer.

Practise 2 Eval, Testing, and Debugging questions →

Model Selection and Optimization 16.8%

  • Pick the smallest model that meets the quality bar, and measure before optimising.
  • Route easy requests to faster models and hard ones to more capable models.
  • Cut cost and latency with caching, batching, shorter context and a sensible max_tokens.
  • Extended thinking helps hard reasoning but adds tokens and latency.

Practise 13 Model Selection and Optimization questions →

Prompt and Context Engineering 11%

  • Write clear instructions and separate inputs with XML-style tags.
  • A few well-chosen examples beat long descriptions.
  • Put long documents before the question and keep only relevant material in context.
  • Summarise or retrieve rather than putting everything into one request.

Practise 9 Prompt and Context Engineering questions →

Security and Safety 8.1%

  • Treat content from users, documents and web pages as untrusted: it can carry prompt injection.
  • Give tools the least privilege they need and require approval for risky actions.
  • Keep secrets out of prompts and logs, and validate outputs before acting on them.
  • Handle personal data according to policy.

Practise 7 Security and Safety questions →

Tools and MCPs 10.6%

  • Write tool descriptions that say what the tool does, its inputs and when not to use it.
  • Use tool_choice to force a call when you need one.
  • MCP servers expose tools, resources and prompts; scope them to what the task needs.

Practise 8 Tools and MCPs questions →

Further practice

For a second set of questions written by someone who has sat the exam, Matthew Purcell publishes an independent Developer – Foundations (CCDV-F) practice set with answers and rationales.

Try three sample questions

Pick your answer, then open the reveal. The full bank has 80 questions.

Sample 1Agents and Workflows
Larkspur Insurance turns each of its 3,000 daily claim forms into a CRM record. Every form goes through the same five steps in the same order. How should you build this?
  1. As a fixed workflow of chained calls, since the steps never change
  2. As an autonomous agent, so it can pick its own path for every form
  3. As a multi-agent system, so each of the five steps can scale alone
  4. As an agent with every tool loaded, ready for new form types later
Reveal answer
Answer: A. When the steps are known in advance and never vary, a fixed workflow is cheaper, faster and easier to test than an agent. Agents earn their cost when the path depends on what they find; scaling steps separately or loading extra tools adds complexity this task does not need.
Sample 2Applications and Integration
Your app at Ashby Travel needs Claude to always reply in a friendly tone and never quote prices. Where do these standing instructions belong in a Messages API request?
  1. In the top-level system parameter, separate from the message list
  2. As the first assistant message, placed before the user ever speaks
  3. In a user message repeated at the end of every request you send
  4. Inside the metadata field, which the model reads before each reply
Reveal answer
Answer: A. The system parameter is the place for role, tone and standing rules that apply to the whole conversation. Faking them as assistant or user turns is less reliable, and the metadata field is not read by the model as instructions.
Sample 3Model Selection and Optimization
Kestrel Mail tags 400,000 support emails a day into one of 12 fixed categories. Accuracy is already high with a mid-tier model, but cost and latency are too high. What should you try first?
  1. Move to the largest model tier and shorten the system prompt
  2. Turn on extended thinking so each label is reasoned through
  3. Evaluate a smaller, faster tier against the same labelled set
  4. Ask the model to explain each label before it gives an answer
Reveal answer
Answer: C. High-volume, simple classification is a natural fit for a smaller, faster tier, and testing it on the labelled set shows whether accuracy holds. A larger tier or extended thinking adds cost and latency, and asking for explanations adds output tokens to every call.

A four-week study plan

  1. Week 1, the API. A third of the exam: messages, tool use, structured output, streaming, errors, caching and batches.
  2. Week 2, agents and models. Workflows versus agents, subagents, and choosing and optimising models.
  3. Week 3, prompts, tools and security. Context engineering, tool and MCP design, prompt injection and least privilege.
  4. Week 4, mock exams. Take the timed mock every other day, drill your weakest domain in between, and aim for 800+.

Frequently asked

How long is the Claude Certified Developer exam?

It is widely reported as 53 questions in 120 minutes, scored from 100 to 1000 with 720 to pass. Check Anthropic's certification pages for the current details.

How much of the exam is about Claude Code?

Very little: about 3%. A third is Applications and Integration, which is mostly the Claude API. For Claude Code itself, see our Claude Code Skills track.

What question formats does it use?

Multiple choice, multiple response (select TWO), Yes/No matching, where you judge each statement in a teammate's notes, and classification, where you sort items into categories. Our mock exam includes all four.

Independent practice material, not affiliated with Anthropic. Questions are original and not taken from any real exam. Always confirm exam details on Anthropic's official certification pages.