Introduction: The Claude Platform and the CCDV-F Exam

The Certification and Its Value

The Claude Certified Developer – Foundations certification (exam code CCDV-F) validates that an individual can build, integrate, and ship production-grade applications, agents, and workflows using Anthropic's Claude platform at a foundational level. Earning it sends a clear signal to employers, clients, and teammates: this person can translate a business requirement into a working system that uses Claude — through API integration, agent and tool construction, prompt and context engineering, evaluation, security, and model selection.

A useful analogy is a private pilot's license: it does not certify that you can fly a fighter jet, but it certifies that you can safely and independently operate an aircraft under normal conditions. The CCDV-F says you can safely take a Claude application from the runway of an idea to the cruising altitude of production. Because the underlying technology evolves rapidly, the credential is valid for 12 months from the date it is awarded.

The exam is targeted at the Minimally Qualified Candidate (MQC) — the theoretical person who has just barely enough skill to be considered competent. The audience is AI/ML engineers, technical leads, and senior software engineers. It is explicitly not for non-technical or casual users, individuals without hands-on development experience, or roles limited to prompt writing alone.

Recommended Experience

Notice how often the word tradeoff appears in the MQC profile. That is a strong hint about the exam's cognitive level: it rarely asks "what is X?" and far more often asks "given these constraints, which of these four reasonable-sounding approaches is best?"

Exam Format and Scoring

A scaled score is not the same as "percent correct." A raw score is converted through a statistical process into a point on the 100–1,000 scale, letting Anthropic keep the difficulty of passing constant even when candidates receive slightly different question sets. Do not walk in assuming you can miss 28% of questions.

The CCDV-F is criterion-referenced: each candidate is measured against a fixed performance standard, not against other candidates. Contrast this with a norm-referenced exam graded on a curve, where only a fixed share passes.

The passing standard was established through a formal standard-setting study, in which trained subject-matter experts judged the performance expected of a minimally qualified candidate. Your score report shows per-domain percent-correct, but those figures help you understand performance — they do not determine pass/fail. Only the total scaled score does, so you cannot "fail" a single domain.

The Eight-Domain Blueprint

Figure 1.1: The eight exam domains grouped by weight tier

Here is where the blueprint becomes a study tool. The naive approach gives every domain equal time; the blueprint tells us that is a mistake. Suppose you have 40 hours to study allocated purely by weight:

Nearly 26 of your 40 hours should go to just the top three domains. That does not mean ignore the small domains — a couple of easy points still count toward 720 — but mastering the Claude API, model selection, and agent construction is where the exam is won or lost. Then adjust up or down based on an honest self-assessment against the MQC profile: the blueprint gives the ideal allocation; your self-assessment personalizes it.

The Claude Platform Landscape

The Claude Agent SDK runs the agent loop inside your own process. It was formerly called the "Claude Code SDK" and renamed to reflect a broader vision: the harness that powers Claude Code can power many other agents. MCP connects agents to external tools and data through a single protocol — MCP is to AI tools what USB-C is to devices, one standard plug replacing a drawer full of proprietary adapters.

Figure 1.3: MCP as one universal protocol between an agent and many external systems

The Model Families: Opus, Sonnet, and Haiku

Note the roughly 5x price gap between Opus and Haiku on output tokens — that gap is precisely why model selection is worth 16.8% of the exam. The most sophisticated production systems use all three tiers together: Haiku routes and handles simple requests, Sonnet does the bulk medium-complexity work, and Opus handles the 10–15% requiring deep reasoning.

Figure 1.2: Multi-tier model routing in a production application

Every layer maps onto an exam domain: the SDK and API to Applications and Integration, the model tiers to Model Selection, the Agent SDK to Agents and Workflows, MCP to Tools and MCPs, and Claude Code to its own domain. Understanding the stack is not separate from passing the exam; it is the exam.