Claude Certified Developer — Foundations
Everything a developer needs to build real applications on Claude — the Messages API, model selection, cost and tokens, and agents — turned into scenario-based flashcards you actually remember with spaced repetition.
- flashcards
- 338
- flashcards
- per day
- ~10 min
- per day
- level
- Beginner → Intermediate
- level
- modules
- 21
- modules
What is the Claude Certified Developer — Foundations certification?
Claude Certified Developer — Foundations (CCDV-F) is Anthropic's entry-level developer certification. It validates a broad, practical understanding of building applications on Claude — how the model and the Messages API actually work, how to choose and optimize a model, and how to design applications and agents — rather than deep expertise in any single feature.
The exam is blueprint-weighted across eight domains, and three of them carry most of the weight: Applications & Integration, Model Selection & Optimization, and Agents & Workflows. The recurring themes are the stateless Messages API and the tool-use loop, matching a model to a task's cost and latency, and knowing when an agent beats a fixed workflow.
This track covers all eight domains, weighted to mirror the blueprint — so most of your study time lands where the exam does — and turns them into scenario-based questions with spaced repetition, moving the mechanics from "I read the docs once" to long-term recall. Every card links to the official Claude documentation for the fact it tests.
21 modules, seed to bloom
Each module is a set of flashcards — 338 in total. Answer, review, and watch your knowledge grow from seed to full bloom.
LLM Fundamentals
How the model works — tokens, context windows, next-token generation, non-determinism, sampling, shot prompting, and fast vs extended thinking
18 cardsTechnical Fundamentals
The integration layer — official SDKs over the REST API, authentication headers, streaming, typed errors, timeouts and retries, and where the SDK ends and the wire begins
15 cardsMessages API Mechanics
Anthropic Messages API — request shape, tool-use cycle, streaming, vision and documents, stop reasons, and usage accounting
18 cardsAPI Optimization: Caching, Batch & Platforms
Cost and latency levers — prompt caching, Batch vs realtime tradeoffs, extended thinking, and calling Claude via Bedrock and Vertex
17 cardsCost & Token Economics
Controlling spend — how input and output tokens drive cost and latency, prompt caching, batch versus real-time pricing, max_tokens, and practical levers to reduce token usage
13 cardsModel Selection Trade-offs
Choosing the right model for the job — balancing capability against latency and cost, matching model tier to task difficulty, and when a smaller or faster model is the correct call
13 cardsSoftware Engineering Foundations
The integration fundamentals under any Claude app — REST, JSON, idempotency, async concurrency, retries, timeouts, and error handling
18 cardsSDLC Integration & Refactoring
Fitting Claude into the development lifecycle — version control flow, code review, CI, refactoring at small and large scale, and safe rollout
16 cardsRequirements & Lifecycle
Turning business needs into an LLM app — functional vs non-functional requirements, constraint-driven choices, scoping, and the operate/maintain lifecycle
18 cardsApplication Design
Architecting apps around Claude — trust boundaries, cross-surface instructions, structured output, session hygiene, and plugins
17 cardsAgent Architecture
What makes a system agentic — the model-driven loop of decide, act, observe; agents versus fixed workflows; when to hand control to the model and when a deterministic pipeline is the safer design
17 cardsAgent Construction
Building an agent in practice — defining tools with clear schemas, the tool_use and tool_result round-trip, who actually executes a tool, handling tool errors in the loop, and least-privilege tool design
18 cardsAgent & Workflow Patterns
Common building blocks — prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer loops, and human-in-the-loop; matching each pattern to the shape of the task
16 cardsContext Engineering
Curating what goes into the window — relevance over volume, compaction and summarization, pruning tool output, isolating subagent context, and fighting context rot in long sessions
19 cardsPrompting Techniques for Claude
Reaching for the right technique — clear direct instructions, XML tags, system prompts, prefilling, examples, and extended thinking, matched to the symptom you are trying to fix
19 cardsModel Context Protocol (MCP)
Connecting Claude to the outside world through a standard — what MCP is, servers and clients, tools/resources/prompts, local vs remote servers, and MCP versus one-off custom tools
18 cardsTools, Built-ins & Skills
Choosing the right capability — platform built-in tools like web search and code execution, custom tools, Agent Skills, and when each fits versus a custom tool or an MCP server
19 cardsLLM Application Security
Securing an app built on Claude — prompt injection and trust boundaries, treating model output as untrusted, jailbreak mitigation, layered guardrails, and validating actions server-side
15 cardsSafety, Keys & Data Handling
Operational safety — protecting API keys, minimizing sensitive data, content moderation, hooks as guardrails, human oversight for high-stakes actions, and handling refusals gracefully
14 cardsClaude Code
Agentic coding in the terminal — what Claude Code is, how it reads and edits a codebase, project memory via CLAUDE.md, permissions, and extending it with MCP, hooks and subagents
11 cardsEvaluation, Testing & Debugging
Measuring quality instead of guessing — defining success criteria, building eval sets, automated and model-graded scoring, catching regressions, and debugging LLM applications systematically
9 cardsSample questions
A taste of the real flashcards. Pick an answer, then reveal the explanation.
A developer is told the model processes text as tokens rather than whole words. What is a token most accurately?
- AA chunk of text such as a word piece or subword, the unit the model reads and generates
- BExactly one English word, since the model always maps a single word to a single token cleanly
- CA single character, since the model reads text strictly one letter at a time as it generates
- DA full sentence, since the model consumes and emits complete sentences as its atomic unit
A team builds a chat feature and assumes the API remembers previous messages by conversation ID, but Claude never recalls earlier turns. Why?
- AThe Messages API is stateless — the client must resend the full conversation history on every request
- BThe Messages API is session-based — a conversation ID must be passed to reload stored server-side turns
- CThe Messages API is cached — earlier turns expire after a fixed idle window and must then be refreshed
- DThe Messages API is streaming-only — history is retained solely while an SSE connection stays open live
A team must pick a model for a new feature. What is the fundamental trade-off across model tiers?
- AMore capable models generally cost more and run slower, so you match the tier to the task's difficulty
- BMore capable models are always cheaper and faster, so the largest model is the right default everywhere
- CAll models cost and perform identically, so the choice among tiers has no effect on the result at all
- DSmaller models are always more accurate, so the least capable tier should be the default for every task
A team debates whether their feature is an agent or a workflow. What most clearly distinguishes an agent?
- AThe model decides which steps and tools to use dynamically, rather than following a fixed path
- BThe agent uses a larger model, since only the top model tier can ever be called an agent at all
- CThe agent runs on a server while a workflow runs only in the browser, which is the sole difference
- DThe agent has no system prompt, since removing all instructions is what makes a system agentic
Learn it once, keep it for good
Answer a question
Each card is one practical concept with multiple options. Pick what you think is right.
Get the full answer
See the correct option plus a clear explanation, and a link to deeper docs when one is available.
Review at the right time
A spaced-repetition engine (SM-2 or FSRS) resurfaces each card just before you would forget it.
Why these foundations are worth your time
Build on Claude with confidence
Knowing the Messages API, tokens and the tool-use loop is the baseline for shipping anything real on Claude — not just passing an exam.
Spend and latency sense
Understanding model trade-offs, prompt caching and batch vs realtime helps you avoid surprise bills and slow apps.
Agent literacy
The agent-vs-workflow decision and the decide-act-observe loop are the fastest-growing skills in AI engineering.
Certification-ready
The cards map directly to all eight CCDV-F exam domains, weighted to match the blueprint.
Common questions
Do I need AI or ML experience first? +
No. The track starts from how a language model and the Messages API work and builds up to agents, so developers new to building on LLMs and those refreshing the basics both benefit. General software engineering familiarity helps.
How long does it take? +
About 10 minutes a day. Spaced repetition means short, frequent sessions beat cramming — most learners get comfortable with the core mechanics within a couple of weeks.
Is it free? +
Yes, completely free. No registration or credit card is required, and all your progress is stored locally in your browser.
Will this prepare me for the CCDV-F exam? +
It covers all eight exam domains, weighted to mirror the official blueprint so most cards land on the heaviest domains (Applications & Integration and Model Selection). It is a companion, not a full course or an exam dump: pair it with the official Anthropic exam guide and hands-on building.
Ready to build on Claude?
Plant your first seed today. Ten minutes a day is all it takes to grow the foundations the CCDV-F exam rewards.