Claude Certified Developer — Foundations

• What it tests
• Read the weights first
• Who it’s for
• The eight domains
• How to prepare
• Common mistakes

What this exam actually tests

CCDV-F certifies that you can build working applications on Claude — agents and workflows with the Agent SDK, integration through the API and client SDKs, custom tools and MCP servers, prompt and context engineering, and the security practices that keep any of it safe to deploy.

It is the most granular of the four Claude blueprints. Where the others publish a weight per domain, this one publishes a weight for each named skill inside each domain, to one decimal place — twenty-five of them. That is unusual, and it makes study planning arithmetic rather than guesswork.

Read the weights before you plan anything

Three numbers reshape how most people would otherwise prepare:

• Applications and Integration is 33.1%. A third of the paper in one domain. Its largest skill alone — Claude Application Design at 8.6% — outweighs five entire domains on this exam.
• Claude Code is 3.1%. On a developer exam. Roughly two items. If you were planning to spend a week in Claude Code because the credential has “Developer” in the title, that week is misallocated.
• Eval, Testing and Debugging is 2.6%. The smallest domain on the paper, and a single skill: debugging and error handling. Notable mostly because it is the opposite of CCAR-P, where evaluation is 16%.

Put together: this exam is far more interested in whether you can build and integrate a Claude application properly than in whether you can operate the tooling around it. Software Engineering Foundations is 7.4% on its own — REST, JSON, async, version control, code review, refactoring — which tells you the bar is general engineering competence applied to Claude, not Claude trivia.

Who it’s for

Anthropic’s recommended profile is one to five years of software engineering, at least six months hands-on with Claude or comparable LLM systems, proficiency in Python and/or TypeScript, fluency with REST APIs and CLI tools, and a working understanding of LLM fundamentals, agents, context management and MCP. There are no mandatory prerequisites — the credential is awarded on exam performance alone.

The honest self-test is whether you have shipped something on the API that other people used. The blueprint assumes you have written tool definitions, handled errors from them, thought about token cost, and had a prompt injection concern at least once.

The eight domains

Skill-level weights are the vendor’s own and are listed in full above this section.

Applications and Integration — 33.1%

Six skills, and the centre of gravity of the whole exam. Claude Application Design (8.6%) covers how Claude interprets instructions across interfaces, content boundaries, schema design, session hygiene and plugin management. Software Engineering Foundations (7.4%) is general practice. Claude API Mechanics (6.8%) is messages, tools, streaming, vision, thinking, caching and the batch-versus-realtime decision. Then Configuration Management (4.1%), Understanding Requirements (3.4%) and Systems Life Cycle (2.8%).

If you allocate study time proportionally, a third of it belongs here.

Model Selection and Optimization — 16.8%

LLM Fundamentals (5.2%) — tokens, context windows, sampling, non-determinism, thinking modes. Technical Fundamentals (6.1%) — SDKs wrapping REST, websockets. Cost and Token Management (2.8%) — budgeting, tracking, prompt caching and cache check-pointing. Model Selection and Tradeoffs (2.7%) — Opus, Sonnet and Haiku against quality, latency and cost, including breaking behaviour changes across releases.

Agents and Workflows — 14.7%

Agent Construction with Claude (5.3%), Agent Patterns and Frameworks (4.9%), Agent Architecture (4.5%). The recurring decision is the first one: workflow or agent. Manager and supervisor hierarchies, subagents, tool-use loops, memory and context-window management, and hooks where you need something deterministic.

Prompt and Context Engineering — 11.0%

Prompt Engineering (4.6%), Context Engineering (3.8%) and Output Handling (2.6%). Output Handling is the one to notice — structured output patterns, response validation, defensive parsing, and what the guide calls scepticism toward confident output. Assume the model will hand you something malformed and design for it.

Tools and MCPs — 10.6%

Tool Implementation (4.4%) — function calling, tool descriptions, error handling, client-side versus server-side, approval patterns. Agentic Customization (4.1%) — choosing among built-in tools, custom tools, Skills and MCPs. MCP Server Development (2.1%) — authoring, deployment, resources and prompts, stdio and socket transports.

Security and Safety — 8.1%

AI Application Security (3.2%) leads: prompt injection, jailbreak defence, untrusted input, data leakage, PII. Then Guardrails and Safe Deployment (2.3%), Identity, Secrets and Key Management (1.6%), and Claude Hooks (1.0%).

Eight percent is more than it sounds on a foundations exam, and prompt injection is the single most likely thing to appear.

Claude Code — 3.1% · Eval, Testing and Debugging — 2.6%

Two items and two items. Claude Code Operation covers rules, skills, commands, agents, memory, session management, the CLAUDE.md hierarchy and settings.json. Debugging and Error Handling covers error typing, recovery strategy, trace analysis, and isolating whether a problem came from the integration layer or the model.

Know them. Do not study them for a week.

How to prepare

Anthropic’s own instruction is to build and operate at least one Claude application that exercises the API, integrates one or more tools, applies prompt and context engineering, and includes basic security and evaluation practice. That one exercise spans the four largest domains — roughly three-quarters of the paper.

• Build the application. Real API calls, at least one custom tool, real error handling, a secret you had to manage properly.
• Then write an MCP server. Small, but end to end — authoring, transport, integration. It converts a 10.6% domain from theory into memory.
• Allocate by weight, not by interest. A third of your time in Applications and Integration. Two hours, not two days, on Claude Code.
• Work the sample questions in section 8 of the exam guide. Published items with full rationale, and the only vendor-calibrated signal available.

Three things candidates get wrong

Assuming a developer exam is a Claude Code exam. It is 3.1%. The paper is about building applications on the API and the Agent SDK.

Skipping the software engineering fundamentals. REST, JSON, async, version control, code review and refactoring are 7.4% in their own right — more than Security and Safety entire. Candidates who came to Claude from a non-engineering background are most exposed here, and it is the least Claude-specific thing on the exam.

Ignoring the decimal places. This blueprint hands you a weight for every one of twenty-five skills. Almost nobody reads them, and they are the most precise study plan any of these four exams gives you.

What Rubric will have for this exam

Mock exams sampled against the published skill weights rather than just the domain weights — which this blueprint makes possible and most competitors ignore. Every answer citing the documentation behind it, dated against the current blueprint, with results reported per domain.

None of that exists yet. This page is the blueprint; the exams follow.