Study Guide

Unofficial practice aid for the Claude Certified Developer – Foundations (CCDV-F) exam. Pair this with Anthropic's official exam guide, Academy materials, and hands-on API work. It is not affiliated with Anthropic.

How to use this guide

1. Skim the domain map and weights.
2. Build small apps that exercise Messages API, tools/MCP, and eval loops.
3. Drill the practice bank (Full Test = 53 questions) and review explanations.
4. Prefer trade-off thinking over trivia memorization.

Domain map (CCDV-F)

Approximate weights used in this practice pack (rounded to sum 100):

1. Applications and Integration

• Stream deltas; assemble content and tool_use blocks before acting. Honor stop_reason.
• Retry 429/5xx with bounded exponential backoff and jitter; respect Retry-After.
• Realtime for interactive chat; Batch for large offline jobs; prompt caching for stable prefixes.
• Prefer schema-constrained structured outputs plus server-side validation.
• Pin model IDs and prompt versions in deployable config; log them per request.
• Cross-cloud: plan for auth, quotas, and feature gaps behind an adapter.

2. Model Selection and Optimization

• Match model tier to latency, cost, and quality.
• Cascades: cheap/fast classifier, escalate hard cases.
• Extended thinking only when reasoning quality is worth the cost.
• Keep cached prefixes byte-identical; put volatile content after the cached section.

3. Agents and Workflows

• Prefer deterministic workflows when steps and validators are known.
• Prefer agents for open-ended exploration with tool feedback loops.
• Correct loop: tool_use -> execute -> tool_result -> continue.
• Cap iterations; detect repeated identical tool failures; escalate cleanly.

4. Prompt and Context Engineering

• Durable policy belongs in versioned system prompts; user turns carry the live task.
• Manage long sessions with summarization/retrieval of salient state.
• Few-shot examples should match production format and include real edge cases.

5. Tools and MCPs

• Typed parameters beat overloaded free-text fields.
• Use tool_choice intentionally: none / auto / any / forced tool.
• Progressive discovery: do not load dozens of unused tools every turn.
• Prefer maintained MCP servers when they fit; build custom for clear gaps.
• Least privilege and clear side-effect labels (read vs write).

6. Security and Safety

• Treat retrieved content as untrusted; do not let docs rewrite system policy.
• Validate tool arguments derived from untrusted text.
• Scope destructive tools by role; require human approval for high impact.
• Redact PII in logs; ACL retrieval so models never see unauthorized rows.

7. Claude Code

• Version-control CLAUDE.md, hooks, and shared slash commands.
• Headless CI: non-interactive flags, vaulted secrets, artifact review.
• Hooks for denylists and egress controls with defense in depth.

8. Eval, Testing, and Debugging

• Keep a versioned regression suite with labels/rubrics beside prompt/model config.
• Score quality and operational budgets (latency/cost).
• Debug with structured traces and redacted offline fixtures.

Practice tips for CCDV-F

• The heaviest domain is Applications and Integration—invest build time in real API streams, retries, and batch jobs.
• Many items reward engineering judgment (workflow vs agent, cache placement, privilege scope).
• After each practice miss, rewrite the correct option in your own words and note the distractor trap.

Good luck—and ship something small to production before exam day.