The 2026 Certification Landscape · lesson 3 of 4
CCDV-F: Developer – Foundations Exam
Format
- Questions: 53.
- Length: 120 minutes (~2:15 per question — the most generous timing of the four exams).
- Fee: $125.
- Passing: 720 on the 100–1,000 scaled score.
- Delivery: Pearson VUE (OnVUE or test center).
- Target profile: AI/ML engineers and tech leads with 1–5 years software engineering plus ~6 months building with Claude.
The eight domains
| # | Domain | Weight |
|---|---|---|
| 1 | Agents and Workflows | 14.7% |
| 2 | Applications and Integration | 33.1% |
| 3 | Claude Code | 3.1% |
| 4 | Eval, Testing and Debugging | 2.6% |
| 5 | Model Selection and Optimization | 16.8% |
| 6 | Prompt and Context Engineering | 11.0% |
| 7 | Security and Safety | 8.1% |
| 8 | Tools and MCPs | 10.6% |
Read the weights strategically: Applications and Integration alone is a third of the exam, and the top three domains (2 + 5 + 1) are ~65%. Claude Code and Eval/Testing together are under 6% — know them, but don't over-invest.
What each domain tests
- Agents and Workflows (14.7%) — agent architecture fundamentals, the agent-sdk, writing custom agent loops, managed agents (Anthropic-hosted, managed sandbox — see claude-platform-101), hooks for deterministic control, and recognizing agentic patterns across frameworks including Strands, LangGraph, and PydanticAI (agentic-patterns).
- Applications and Integration (33.1%) — the core third:
- API mechanics — messages-api request/response, tools, streaming (streaming), vision (vision), extended thinking (extended-thinking), prompt caching (prompt-caching), batch (batch-api).
- Software engineering foundations — error handling, retries, idempotency applied to LLM calls.
- Application design — where Claude sits in a system; sync vs async; state ownership.
- Configuration management — CLAUDE.md, settings.json, and model pinning (specifying exact model IDs so behavior doesn't shift under you when aliases move).
- Claude Code (3.1%) — rules/skills/commands/agents/memory, headless mode (non-interactive CI use), and the CLAUDE.md hierarchy (claude-code-md, claude-code-overview).
- Eval, Testing and Debugging (2.6%) — basic evaluation harnesses, regression-testing prompts, debugging failed generations.
- Model Selection and Optimization (16.8%) — tokens and context windows, sampling parameters, fast mode, extended/adaptive thinking and effort levels (dialing reasoning depth per request); cost/token management with prompt caching as the flagship lever.
- Prompt and Context Engineering (11.0%) — preventing drift (behavior sliding over long sessions) and bloat (context filling with low-value tokens), compaction, subagent isolation (giving noisy subtasks their own context — subagents); few-shot prompting (few-shot-prompting); output handling and defensive parsing — never trusting model output to be well-formed; validate and repair (structured-output).
- Security and Safety (8.1%) — prompt injection (untrusted content steering the model) and jailbreak defense, PII handling, guardrail layering (multiple independent defenses rather than one filter), least privilege for tools, hooks as deterministic enforcement, and secrets management (never in prompts or CLAUDE.md).
- Tools and MCPs (10.6%) — function calling mechanics (tool-use), client-side vs server-side tools (who executes: your code vs Anthropic's infrastructure), MCP server development (mcp-server-python), and the tradeoff matrix of built-in Tools vs custom Tools vs Skills vs MCPs — roughly: built-in for commodity capabilities, custom tools for app-specific actions, Skills for reusable know-how, MCP for shareable cross-surface integrations.
Study plan against the weights
- Weeks 1–2: Domain 2 — work through messages-api, tool-use, streaming, prompt-caching, extended-thinking; build one small app that uses all five.
- Week 3: Domains 5 + 6 — token economics, thinking/effort controls, context-management, defensive parsing.
- Week 4: Domains 1 + 8 — agent-sdk, agentic-patterns, one toy MCP server; then sweep 3/4/7 with claude-code-overview and the security bullets above.
See also
- certifications-overview — logistics, retakes, renewal shared by all exams
- claude-platform-101 — the course mapped to this exam
- cca-f-exam-overview — the architect-track sibling (CCAR-F)