Claude Certification Exam Prep

Claude Certification Exam Prep (2026)

Study materials, hands-on labs, and a fully offline study app for Anthropic's Claude certifications. The repo is built to hold any number of exams: content lives in reusable, versioned components that each exam pins, and the study app, CLI, and TUI switch between exams.

CCAR-F validates practical judgment about building production applications with Claude Code, the Claude Agent SDK, the Claude API, and the Model Context Protocol (MCP). CCDV-F validates building, integrating, and shipping Claude-powered applications, agents, and workflows — API integration, agent and tool construction, prompt and context engineering, evaluation, security, and model selection. This repo pairs each exam's official guide with two ways to prepare: a buildable, testable lab curriculum and a polished, fully offline study app (a three-tier practice quiz with real "Select N" multi-response items, flashcards, concept explainers, browsable labs, and a readiness dashboard) — so you gain hands-on experience with every feature, concept, and topic the exams cover, not just read about them.

By the Numbers

Both exams draw from one shared content pool (639 questions, 196 flashcards, 55 concept explainers; 150 items serve both exams), each through its own item map. Questions span three difficulty tiers — CCAR-F: Standard (240), Intermediate (120), and Hard (75, including 10 multi-response "Select N" items calibrated to the real exam's format); CCDV-F: Standard (192), Intermediate (65), and Hard (65, including 18 "Select N") — dynamically selectable in the Web UI, or chosen with --set (any combination) in the CLI/TUI. CCAR-F covers all 5 exam domains and 30 task statements (40 syntax-highlighted code samples across its concept explainers); CCDV-F covers all 8 domains and 25 skills, with Standard-tier questions per skill scaled to the skill's exam weight (24 code samples). Questions and concepts link to the lab that drills them where one exists. The whole study app runs locally and offline, with zero runtime dependencies.

Study Materials

Labs At A Glance

• 25 labs (written for CCAR-F; CCDV-F links the 19 that fit its skills) across 5 dependency tiers (Foundations → Capstones), ordered so prerequisites come first, then by difficulty (1–10). ~48 hours of hands-on effort.
• Full CCAR-F coverage: every task statement (1.1–5.6) and every exam scenario (S1–S6) maps to at least one lab — see the traceability tables in labs/README.md.
• Each lab folder contains: README.md (instructions), SOLUTION.md (reference key + why-the-distractors-are-wrong), starter/ (scaffold you fill in), solution/ (reference implementation), tests/ (automated checks), and any config artifacts (.claude/rules, .mcp.json, SKILL.md, CI YAML, sample docs).
• Tests are deterministic and offline by default: labs mock the Claude API via labs/_shared/mock_anthropic.py, so a plain test run needs no API key and costs nothing. A handful of inherently semantic checks are marked llm and run only when you opt in with an API key.

Prerequisites

• uv — drives all Python scripts and tests (no manual venv needed)
• Claude Code — for the configuration-oriented labs
• Optional: an ANTHROPIC_API_KEY to exercise the live end-to-end and -m llm paths

Quick Start

First install uv — macOS/Linux: curl -LsSf https://astral.sh/uv/install.sh | sh; Windows (PowerShell): irm https://astral.sh/uv/install.ps1 | iex (full guide).

git clone https://github.com/craigforr/cert-anthropic-2026
cd cert-anthropic-2026

# The study app (quiz, flashcards, concepts, labs, readiness):
uv run study/serve.py                     # opens http://localhost:8000/web/

# The labs:
cd labs
uv run pytest lab-03-agentic-loop        # test YOUR work on one lab (its starter/)
uv run pytest                            # test your work across all labs
LAB_TARGET=solution uv run pytest        # run the reference solutions (all green)
ANTHROPIC_API_KEY=sk-... uv run pytest -m llm   # add the optional semantic checks

Suggested workflow per lab: read its README.md (note prerequisites) → fill in starter/ until uv run pytest lab-NN-... passes → compare against SOLUTION.md and solution/ for the why, not just the what.

Study Path

Work the labs in the order given in labs/README.md (dependency tier, then difficulty). Tier 0–1 build the core primitives (agentic loop, tool use, structured output, Claude Code config); Tier 2–3 layer on error handling, orchestration, and context management; the Tier 4 capstones integrate everything into the six CCAR-F exam scenarios.

Notes

• Certification content is derived from the publicly distributed exam guides.
• Test suite status: labs (cd labs && LAB_TARGET=solution uv run pytest) → 376 passed, 3 llm-marked tests deselected (they run only with an API key); study content and app (cd study && uv run pytest) → 71 passed.