The exam

Claude Certified Architect
Foundations (CCA-F)

The first AI architecture certification that tests whether you can actually build production systems with Claude. Not trivia. Scenarios.

60
Questions
120
Minutes
720
To pass
$99
Exam fee

How the exam works

The CCA-F is scenario-based. The exam has 6 built-in scenarios (real-world production situations). When you sit the exam, 4 of the 6 are randomly selected. Every question is anchored to one of those scenarios.

You're not memorizing definitions. You're dropped into a situation -- a customer service bot, a medical triage system, a code review pipeline -- and asked to make the right architectural call.

The exam is proctored. You cannot have Claude open during the exam. Scored on a 100-1,000 scale. 720 to pass. No penalty for guessing.

The #1 tested concept: programmatic enforcement vs prompt-based guidance. When the stakes are high (money, safety, access), the answer is code, not prompts. This single pattern covers 27% of your score.

Exam weight by domain
27%
18%
20%
20%
15%
D1
D2
D3
D4
D5
The 5 domains
Domain 1

Agentic Architecture & Orchestration

27%

How to design AI agent systems that actually work in production. Agent loops, task decomposition, multi-agent communication, failure handling, and orchestration patterns.

KEY INSIGHT
When to enforce rules with code vs prompts. If patient safety, financial controls, or access permissions are at stake, the answer is always programmatic enforcement.
EXAM TRAP
Thinking "add it to the system prompt" is good enough for critical guardrails. It's not. Prompts are suggestions. Code is law.
What's tested
-Agent loop design and stop conditions
-Task decomposition strategies
-Inter-agent communication patterns
-Failure handling and graceful degradation
-Programmatic enforcement vs prompt guidance
-Human-in-the-loop placement
Domain 2

Tool Design & MCP Integration

18%

How to give Claude access to the outside world. Function calling, MCP server architecture, tool schema design, error handling, and external API integration.

KEY INSIGHT
Tool use and function calling are heavily tested. You need to know how to design tool schemas, handle errors, and integrate external APIs cleanly.
EXAM TRAP
Building overly complex tool chains when a simple function call would suffice. Keep tools atomic and composable.
What's tested
-Function calling patterns
-MCP server architecture
-Tool schema design (JSON Schema)
-Error handling for external tools
-API integration patterns
-Tool composition and chaining
Domain 3

Prompt Engineering & Structured Output

20%

Architecture-level prompt design. Not "write a better prompt" but "design scalable prompt systems." System prompts, structured output schemas, injection defense, chain-of-thought.

KEY INSIGHT
This is not basic prompt writing. It's designing prompt systems that scale across an entire application. XML tags, JSON schemas, few-shot patterns, and defense against injection.
EXAM TRAP
Confusing "prompt engineering" with casual chatting. The exam tests architectural decisions about prompt systems, not clever wording.
What's tested
-System prompt architecture
-XML and JSON output schemas
-Few-shot pattern design
-Prompt injection defense
-Chain-of-thought strategies
-Structured output validation
Domain 4

Claude Code Configuration & Workflows

20%

How to configure Claude Code for real development workflows. Configuration files, CI/CD integration, DevOps tooling, and workflow automation.

KEY INSIGHT
Know the configuration hierarchy, how CLAUDE.md files work, permission models, and how to integrate Claude Code into existing development workflows.
EXAM TRAP
Ignoring the configuration system. Claude Code is highly configurable and the exam expects you to know how to set it up for team workflows.
What's tested
-CLAUDE.md and configuration hierarchy
-Permission models and security
-CI/CD pipeline integration
-Workflow automation patterns
-Team configuration and sharing
-Custom slash commands and hooks
Domain 5

Context Management & Reliability

15%

How to manage Claude's context window effectively and build reliable systems. Token budgeting, context pruning, caching strategies, and graceful degradation.

KEY INSIGHT
Understanding token limits, when to summarize vs cache, and how to design systems that degrade gracefully when context runs out.
EXAM TRAP
Assuming infinite context. Every token costs money and attention. The exam tests whether you can design systems that work within real constraints.
What's tested
-Token budgeting and optimization
-Context window management
-Caching and prompt caching
-Graceful degradation patterns
-Human-in-the-loop workflows
-Reliability and error recovery
The PencilPrep path

4 levels. 125 lessons. Zero to exam-ready.

Every other study guide starts at Level 2 and assumes you already know what an API is. PencilPrep starts at Level 0 and assumes you know nothing.

LEVEL 0LIVE

Absolute Zero

Computing fundamentals. What is a computer, what is data, what is JSON. You start here if you have zero technical background.

33
lessons
7 weeks
Start Level 0
LEVEL 1

Foundation Concepts

APIs, HTTP, tokens, models, and the basics of how Claude works under the hood. The bridge from general computing to AI-specific knowledge.

30
lessons
6 weeks
LEVEL 2

Architecture Patterns

Prompt engineering, tool design, MCP integration, and agent orchestration. The core of the CCA-F exam. Every domain is covered here.

35
lessons
7 weeks
LEVEL 3

Exam Ready

Scenario-based practice, exam patterns, trap answers, and timed drills. You finish here ready to sit the exam with confidence.

27
lessons
5 weeks
Who is this for

Product managers who need to architect Claude into their products without depending on engineering for every decision.

Developers new to AI who want to build production Claude systems, not just chat with it.

Solutions architects evaluating Claude for enterprise deployments.

Career changers who want a credential that proves they can build with AI, not just talk about it.

Frequently asked

Can I pass the CCA-F with no technical background?

Yes. That's literally why PencilPrep exists. Level 0 teaches you what a computer is. Level 3 has you answering exam-level scenario questions. The path is 125 lessons long, but every step is designed for someone starting from zero.

How long does it take to go from zero to CCA-F certified?

At one lesson per day (10 minutes), the full curriculum takes about 25 weeks. If you already have a technical background, you can skip Level 0 and finish in about 18 weeks. Experienced developers can skip to Level 2 and finish in 12 weeks.

What do I need to know before starting?

Nothing. If you can read this sentence and you have an internet connection, you're qualified to start. That's the point.

Is the CCA-F exam hard?

It's scenario-based, not trivia-based. You can't memorize your way through it. But the concepts are learnable by anyone. The pass rate isn't published, but the exam rewards understanding patterns, not recalling facts.

Is PencilPrep free?

Every lesson is free. Forever. No paywalls, no premium tier, no "upgrade to see the answer." The exam itself costs $99 and is administered by Anthropic.

Ready to start from zero?

Get the first CCA-F lesson by text. One concept, 10 minutes, no experience required.

1 text/day. Unsubscribe anytime.
Start Level 0 now