CCDV-F study guide: domains, format and a 4-week plan

Claude Certified Developer: Foundations (CCDV-F) is Anthropic's certification for engineers who build, integrate and ship applications, agents and workflows on Claude. It covers the Claude API, the Claude Agent SDK, Claude Code, custom tools and MCP servers, prompt and context engineering, and security. You answer 53 multiple-choice and multiple-response questions in 120 minutes, proctored online or at a Pearson VUE test centre.

CCDV-F at a glance

Details as of September 2026 — confirm on the official exam page before booking.

With 53 items in two hours you have a little over two minutes per question, more than on the other Claude exams. The credential is valid for 12 months, with a free on-time renewal assessment on the Anthropic Partner Academy.

Who this exam is for

The exam guide targets AI and machine learning engineers, technical leads and senior software engineers. Its recommended experience is one to five years of software engineering, at least six months hands-on with Claude or a comparable LLM system, proficiency in Python or TypeScript, and fluency with REST APIs and command-line tools. None of this is mandatory; the credential is awarded on exam performance alone. The guide excludes non-technical users and roles limited to prompt writing.

Anthropic's certification FAQ states that exams are open to people at Claude Partner Network organisations, registering with a partner email address on a recognised company domain. Confirm your eligibility before planning.

How CCDV-F differs from the other Claude certifications. The four exams are separate credentials with no formal prerequisites.

• [Associate: Foundations (CCAO-F)](/blog/ccao-f-study-guide) is for professionals using Claude through its apps, with no coding.
• Developer: Foundations (this exam) is the builder's exam. It is the broadest of the four on general software engineering: REST, JSON, asynchronous code, version control and the systems life cycle all appear, alongside Claude-specific API mechanics.
• [Architect: Foundations (CCAR-F)](/blog/ccar-f-study-guide) uses the same technologies but asks for design trade-offs inside four production scenarios, with heavier weight on agent orchestration and Claude Code configuration.
• [Architect: Professional (CCAR-P)](/blog/ccar-p-study-guide) moves up to enterprise solution design, retrieval, governance, compliance and stakeholder communication.

What the exam covers

The blueprint has eight domains, each split into weighted skills. Two are much smaller than their names suggest.

Agents and Workflows (14.7%)

When a fixed workflow beats an autonomous agent, how supervisor and subagent hierarchies divide work, and how agents are built: the Agent SDK, custom agent loops, self-hosted versus Anthropic-hosted deployment, and hooks for deterministic actions. The guide also names third-party frameworks (Strands, LangGraph, PydanticAI), so know what an abstraction framework gives you and what it hides.

Applications and Integration (33.1%)

A third of the exam. It spans requirements, the systems life cycle, Claude API mechanics (messages, tools, streaming, vision, thinking, caching, batch versus real-time, access through cloud providers), general software engineering, application design and configuration management. Application design (8.6%) is the largest single skill: how Claude interprets instructions across Claude Code, the desktop and web apps, the API and SDKs, plus schema design, session hygiene and plugins. Configuration covers CLAUDE.md, settings.json, model version pinning and prompt versioning.

Claude Code (3.1%)

Small as a domain, though Claude Code also appears under application design and configuration. Know rules, skills, commands, subagents and memory, the CLAUDE.md hierarchy, headless mode for automation, session management and settings.json.

Eval, Testing, and Debugging (2.6%)

Despite the name, the blueprint's only skill here is debugging and error handling: identifying error types, choosing a recovery strategy, reading traces, and working out whether a fault lies in your integration code or in the model's output.

Model Selection and Optimization (16.8%)

LLM fundamentals (tokens, context windows, sampling, non-determinism), thinking and effort options, SDKs over REST and WebSockets, choosing between the Opus, Sonnet and Haiku tiers on quality, latency and cost, behaviour changes between model releases, and token and caching strategy for cost. Learn the reasoning, not specific prices or limits, which change.

Prompt and Context Engineering (11%)

Keeping context lean (pruning tool output, compaction, isolating work in subagents), prompt placement between system and user turns, few-shot examples, input sanitisation, and treating output defensively: validate structure, parse carefully and stay sceptical of confident answers.

Security and Safety (8.1%)

Prompt injection, jailbreaks, untrusted input, data leakage and PII; guardrail layering and least privilege; hooks that block destructive actions; and handling API keys and secrets across environments.

Tools and MCPs (10.6%)

Writing tool definitions and descriptions, client-side versus server-side tools, approval patterns, building MCP servers (tools, resources and prompts, and their transports), and choosing between built-in tools, custom tools, Skills and MCP for a given job.

A 4-week study plan

Time follows the weights, so Applications and Integration gets the most attention.

Week 1: Applications and Integration, part one. Build a small service that calls the Messages API with streaming, one tool, an image input and prompt caching. Then run the same workload through the batch interface. Revise REST, JSON and async error handling.

Week 2: Applications and Integration, part two, plus Claude Code. Set up a repository with a project CLAUDE.md, settings.json and one skill, run Claude Code headless in a script, and pin a model version in configuration. Practise explaining why the same instruction behaves differently in Claude Code, the API and the web app.

Week 3: Models, agents and context (about 43% together). Build an agent with the Agent SDK that uses a subagent for a noisy search step, and add a hook. Compare two model tiers on one task for quality, latency and token use. Add few-shot examples and schema validation, and watch what changes.

Week 4: Tools, security, debugging and rehearsal. Write a minimal MCP server with one tool and one resource. Add a hook that blocks a destructive command, and feed your agent a document containing an injected instruction to see what your defences catch. Sit a timed mock and review every miss.

Where candidates lose marks

• Studying by domain name. Claude Code and Eval are only 3.1% and 2.6% as domains, while Applications and Integration is 33.1%. Candidates who spend a week on Claude Code features and skim REST and async fundamentals have the ratio backwards.
• Agent when a workflow would do. If the steps are known and fixed, a scripted workflow is cheaper and more predictable. Answers that add autonomy without need are common distractors.
• Prompts where hooks belong. When an action must never happen, such as deleting production data, a prompt instruction is guidance and a hook is enforcement. The guide's security skills point to the deterministic control.
• Blaming the model for integration bugs. Debugging questions often describe a symptom that looks like a bad answer but comes from truncated context, a malformed tool result or a parsing error. Isolate the layer first.
• Outdated vocabulary. Some guide terms have moved in current documentation: Claude Code has folded custom commands into skills, and MCP's standard transports are stdio and Streamable HTTP. Know the guide's wording and today's names.
• Multiple-response items. Each item states how many answers to select. Judge each option separately.

Practising for CCDV-F on CertifyAIx

CertifyAIx offers original practice questions written against Anthropic's published exam guide and weighted by skill. The [CCDV-F practice questions](/certifications/claude-certified-developer-foundations) page includes:

• 634 original questions across all eight domains
• 3 full-length mock exams with 53 questions in 120 minutes, the same as the real exam
• 29 mixed domain papers and 33 topic papers of up to 25 questions each (60 minutes per 25)
• An explanation for every question, covering why each option is right or wrong

See [how CertifyAIx practice questions are written](/blog/how-certifyaix-practice-questions-are-written) and [how to use practice exams effectively](/blog/how-to-use-practice-exams-effectively).

Frequently asked questions

How hard is CCDV-F?

It assumes working software engineering experience and real time spent building with Claude or a similar model. The breadth across general engineering topics is what usually catches people out.

How long should I prepare?

Three to six weeks is a sensible range. The shorter end fits an engineer already building agents on Claude; the longer end allows time to get hands-on with the Agent SDK, Claude Code and MCP if you have used only the raw API.

Are CertifyAIx questions real exam questions?

No. Every CertifyAIx question is original. Anthropic treats exam content as confidential under its non-disclosure agreement, and misconduct can cost you the credential.

What is the retake policy?

After a failed attempt you wait 14 days, then 30 days after a second failure and 90 days after a third, with up to four attempts in a rolling 12 months. Each attempt is charged at the full fee.

Should I take CCDV-F or CCAR-F?

Take CCDV-F if your work is writing and debugging Claude integrations. CCAR-F suits people who make architectural choices about agents, tools and Claude Code across a team, and its questions are organised around production scenarios.

CertifyAIx is independent and not affiliated with Anthropic. Claude Certified Developer: Foundations is a trademark of its owner. All CertifyAIx practice questions are original.