Anthropic launched the Claude Certified Architect Foundation (CCAF) certification โ the first vendor-backed credential for professionals designing production systems on Claude’s platform. Whether you’re building agentic workflows, integrating tools through MCP, or architecting reliable AI pipelines, this Claude AI certification validates the skills that matter.
I put together a comprehensive CCAF study guide while preparing for the exam, and I’m releasing it as an open-source resource. Here’s what’s in it and why it’s worth your time if you’re pursuing Claude certification.
What the CCAF Exam Covers
The Claude Certified Architect Foundation exam spans five weighted domains:
| Domain | Weight | Focus |
|---|---|---|
| Agentic Architecture & Orchestration | 27% | Designing agent loops, multi-agent patterns, session management |
| Claude Code Configuration & Workflows | 20% | CLAUDE.md hierarchy, skills/commands, CI/CD integration |
| Prompt Engineering & Structured Output | 20% | JSON schemas, few-shot prompting, data extraction |
| Tool Design & MCP Integration | 18% | MCP server architecture, tool definitions, error handling |
| Context Management & Reliability | 15% | Context windows, caching, compaction, retry strategies |
The heaviest domain โ Agentic Architecture โ is also the most practically relevant. If you’re building anything beyond single-turn Claude interactions, you need to understand how loops, sessions, and subagents work at the platform level. Drawing from our experience building secure agentic systems for enterprise clients, this guide distills both exam knowledge and practical production patterns.
What’s in the CCAF Study Guide
The guide is organized by domain in recommended reading order. Each domain includes:
- A detailed study guide covering key concepts, architectural patterns, reference diagrams, pseudocode, and exam-relevant checklists
- 10 practice questions per domain (50 total) in the style of the actual exam โ single best answer, multiple choice
- An answer key with rationale explaining why each answer is correct โ not just which letter to pick
The material draws from Anthropic’s official documentation, the Claude API reference, official cookbooks, and primary research on agentic systems (ReAct, Reflexion, AutoGen, etc.).
CCAF Exam Topics & Key Study Areas
Agentic Architecture & Orchestration (Domain 1 โ 27%)
This is the largest domain and goes deep on the tool_use / tool_result contract, the difference between building your own loop on the Messages API versus using the Agent SDK, subagent isolation patterns, and multi-agent coordination protocols including contract net and blackboard architectures. There’s a full reference architecture with Mermaid diagrams and pseudocode for a robust ReAct-style loop.
Claude Code Configuration & Workflows (Domain 2 โ 20%)
Covers the CLAUDE.md hierarchy from managed policy down to lazy-loaded subdirectory files, the new Skills system (replacing legacy commands), permission modes, sandboxing, and CI/CD pipelines for GitHub Actions, GitLab, Jenkins, and CircleCI. It includes a working config validator skill you can actually use in your projects.
Prompt Engineering & Structured Output (Domain 3 โ 20%)
Focuses on what actually works for Claude exam prep: XML tag delimiting, few-shot examples, the distinction between JSON outputs (output_config.format) and strict tool use (strict: true), and common exam traps like the system parameter placement and outdated thinking configuration. Covers adaptive thinking with the effort parameter for Claude Opus 4.6 and Sonnet 4.6.
Tool Design & MCP Integration (Domain 4 โ 18%)
Covers the Model Context Protocol roles (host, client, server), transport types (HTTP, SSE, stdio), configuration scoping (local, project, user), and โ most importantly โ tool design best practices that make Claude choose and use your tools correctly. Single-purpose tools, routing-quality descriptions, tight schemas, idempotency, and the critical difference between tools, resources, and prompts.
Context Management & Reliability (Domain 5 โ 15%)
Brings it all together: prompt caching (5-minute and 1-hour TTL), server-side compaction, context editing (tool-result and thinking-block clearing), retry strategies by error class (429 vs 529 vs 5xx), budget controls (max_turns, max_budget_usd, fallback-model), permission modes, and human-in-the-loop escalation patterns.
Who This CCAF Study Guide Is For
- Developers building production applications on Claude’s APIs
- Architects designing multi-agent systems and tool integrations
- Engineers preparing specifically for the CCAF certification exam
- Anyone who wants a structured, no-fluff reference for Claude’s platform capabilities
Whether you’re pursuing the Claude AI certification for career growth or just want to deepen your Claude architecture knowledge, this guide covers what you need.
CCAF Exam FAQ
Is the CCAF certification worth it?
If you’re building production systems on Claude, yes. It’s the first vendor-backed credential that validates real architectural skills โ agentic loop design, MCP integration, context management, and reliability engineering. It signals to employers and clients that you can design Claude-based systems that actually work at scale.
What are the prerequisites?
Anthropic recommends hands-on experience with Claude’s APIs and familiarity with agentic patterns. There’s no formal prerequisite, but you’ll get the most value if you’ve built at least one tool-using Claude application. Check Anthropic’s certification page for the latest requirements.
How do I register, and what does the CCAF exam cost?
Registration is through Anthropic’s certification portal. Pricing may change, so check the official page for current exam cost details.
What’s the exam format?
The CCAF exam is multiple-choice with single best answer questions across the five weighted domains. See Anthropic’s official exam blueprint for the latest details on question count, time limit, and passing score.
How difficult is the CCAF exam?
It tests practical architectural knowledge, not trivia. If you understand how tool_use contracts work, can design a multi-agent orchestration pattern, know the CLAUDE.md hierarchy, and can reason about context budgets โ you’re in good shape. The practice questions in this guide are calibrated to match exam difficulty.
Any tips for exam day?
Read questions carefully โ many distractors are plausible but subtly wrong. Watch for outdated API patterns (e.g., thinking=true is incorrect for Claude 4.6, system is a top-level parameter not a message role). Budget your time across domains proportional to their weight.
Get the Study Guide
The entire CCAF study guide is open source on GitHub:
github.com/kkaminsk/ClaudeStudyGuide
Clone it, fork it, study from it. If you find errors or want to contribute improvements, PRs are welcome. Join the GitHub Discussions to connect with other CCAF candidates, ask questions, and share study strategies.
Need help architecting your next AI solution? Big Hat Group builds production Claude-based systems โ agentic workflows, MCP integrations, and enterprise AI platforms. Explore our consulting services or get in touch.