Anthropic Claude Certified Architect – Foundations (CCAO-F) Exam Preparation Hub

📌 Executive Exam Overview

The Claude Certified Architect – Foundations (CCAO-F) is the official architecture accreditation from Anthropic. This certification validates that an enterprise architect or AI engineer can scope, architect, and deploy production-grade solutions powered by Claude. It benchmarks competence across model selection, deployment platform evaluation (Anthropic API, Amazon Bedrock, Google Cloud Vertex AI), agentic vs. single-shot architectures, Model Context Protocol (MCP) integration, developer tooling (Claude Code), context management economics, prompt engineering, and responsible AI guardrails.

🎯 Official Exam Specification Matrix

📊 Exam Domain Breakdown & Weightings

┌─────────────────────────────────────────────────────────────────────────────────┐
│                    CCAO-F DOMAIN WEIGHT DISTRIBUTION                            │
├──────────────────────────────────────────────────────┬──────────┬───────────────┤
│ Domain Area                                          │ Weight   │ Est. Questions│
├──────────────────────────────────────────────────────┼──────────┼───────────────┤
│ 1. Agentic Architecture & Orchestration              │   27%    │  ~16 Questions│
│ 2. Tool Design & Model Context Protocol (MCP)        │   18%    │  ~11 Questions│
│ 3. Claude Code Configuration & Developer Workflows   │   20%    │  ~12 Questions│
│ 4. Prompt Engineering & Structured Output            │   20%    │  ~12 Questions│
│ 5. Context Management & Production Reliability       │   15%    │  ~9 Questions │
├──────────────────────────────────────────────────────┴──────────┴───────────────┤
│ Total: 100% | 60 Questions | 120 Minutes                                        │
└─────────────────────────────────────────────────────────────────────────────────┘

pie title CCAO-F Exam Domain Weightings
    "Domain 1: Agentic Architecture & Orchestration (27%)" : 27
    "Domain 2: Tool Design & MCP Integration (18%)" : 18
    "Domain 3: Claude Code & Workflows (20%)" : 20
    "Domain 4: Prompt Engineering & Structured Output (20%)" : 20
    "Domain 5: Context Management & Reliability (15%)" : 15

🗂️ Repository Architecture & Navigation Index

Explore the comprehensive study and architecture modules in this repository:

🎯 Featured Demo Practice Questions (Front Page Previews)

Below are representative scenario questions highlighting each of the 5 official exam domains. Test your understanding, then click to expand the full architectural breakdown.

(For the complete bank of 25 questions, see DEMO_QUESTIONS.md.)

🔹 Demo 1: Agentic Architecture & Orchestration (Domain 1 — 27%)

Topic: Agent Loop Control & Runaway Prevention

Scenario:
An AI Architect is designing a customer support agent using Claude 3.5 Sonnet that uses tools to fetch account details, check transaction statuses, and issue refunds. During load testing, the agent occasionally enters an infinite loop when an external payment API returns a generic HTTP 500 error, continuously calling the tool with identical arguments until hitting system timeouts.

Question:
Which architectural pattern should the architect implement to most reliably prevent this runaway behavior while maintaining system resilience?

• A) Implement an exponential backoff retry loop directly inside the prompt instructions by telling Claude to wait 5 seconds before retrying.
• B) Wrap the tool execution in the application runtime with loop detection, enforce a deterministic max_iterations counter, and return a structured error status in tool_result upon repeated failures.
• C) Switch the underlying model from Claude 3.5 Sonnet to Claude 3.5 Haiku, as Haiku has lower latency and smaller agent loops.
• D) Remove the payment API tool from the agent and require all payment queries to be answered using few-shot prompt examples.

🔹 Demo 2: Tool Design & MCP Integration (Domain 2 — 18%)

Topic: Model Context Protocol (MCP) Transports & Remote Cloud Access

Scenario:
An organization wants to build an internal MCP server that allows developer workstations running Claude Code to query an internal PostgreSQL database located in a private AWS VPC. Developers work remotely from various corporate laptops.

Question:
Which MCP transport and architecture should the architect select?

• A) stdio transport running directly over an open public internet port.
• B) SSE (Server-Sent Events) transport hosted on an authenticated internal container behind an enterprise VPN / API Gateway with TLS encryption.
• C) WebSockets without TLS authentication.
• D) Embed the raw database credentials in a static markdown file on each developer's desktop.

🔹 Demo 3: Claude Code Configuration & Workflows (Domain 3 — 20%)

Topic: Automated CI/CD Headless Execution

Scenario:
A DevOps engineer wants to integrate Claude Code into a GitHub Actions pipeline to automatically analyze code changes on pull requests, run linter checks, and comment recommendations.

Question:
How should the engineer configure the Claude Code CLI command for automated, non-interactive execution in CI/CD?

• A) claude --interactive --auto-confirm
• B) claude --non-interactive -p "Analyze git diff against main, run lint, and summarize findings" with authenticated ANTHROPIC_API_KEY stored in repository secrets.
• C) claude run --gui-mode --headless
• D) Pipe standard input from an infinite bash yes loop into claude.

🔹 Demo 4: Prompt Engineering & Structured Output (Domain 4 — 20%)

Topic: Assistant Message Prefilling for Structured JSON Output

Scenario:
An application requires Claude to output raw JSON adhering to a strict schema: {"status": "APPROVED" | "REJECTED", "confidence": float, "reason": string}. However, Claude occasionally includes conversational filler before the JSON (e.g., "Here is the evaluation JSON:"), which causes the downstream JSON parser to crash.

Question:
What is the most robust and token-efficient method to guarantee Claude outputs ONLY valid JSON without preamble?

• A) Add "DO NOT INCLUDE PREAMBLE OR CONVERSATION" 10 times in the prompt.
• B) In the Messages API call, prefill the assistant message with { or {"status": to force Claude to continue directly with the JSON payload.
• C) Parse the output with regular expressions and guess missing brackets.
• D) Switch to a model that does not support conversational text.

🔹 Demo 5: Context Management & Reliability (Domain 5 — 15%)

Topic: Prompt Caching Strategy & Cost Optimization

Scenario:
A SaaS company provides a coding assistant for a codebase containing 40,000 tokens of documentation and boilerplate libraries. Thousands of users query the assistant every minute with short 50-token questions. Without caching, API costs are unsustainable.

Question:
How should the architect implement Anthropic Prompt Caching to achieve the maximum possible cost and latency reduction?

• A) Place a cache_control: {"type": "ephemeral"} breakpoint at the end of the static 40,000-token documentation and system prompt block, placing dynamic user questions at the end of the messages array.
• B) Place the dynamic user query at the beginning of the prompt and the 40,000-token documentation at the very end.
• C) Enable caching on every single user turn regardless of token length.
• D) Cache only outputs generated by the model.

👉 Want to practice all 25 questions? Check out the full question bank in DEMO_QUESTIONS.md.

🚀 Recommended Preparation Resources & Practice Tests

1. Official Anthropic Academy Learning Path

Anthropic provides free curriculum courses designed to prepare architects for the CCAO-F credential:

• 🎓 AI Fluency: Framework & Foundations (Level 100): Ethical AI, system safety, collaboration fundamentals.
• 🎓 Claude 101 (Level 100): Core Claude capabilities, artifacts, Projects, and interface workflows.
• 🎓 Building with the Claude API (Level 100–200): Authentication, Messages API, system prompts, tool use, streaming, and evaluations.
• 🎓 Claude with Amazon Bedrock (Level 100–200): IAM security, Bedrock Converse API, VPC endpoints, Provisioned Throughput.
• 🎓 Claude on Google Cloud (Level 100–200): Vertex AI Model Garden, Service Accounts, Private Service Connect, regional quotas.
• 🎓 Introduction to Model Context Protocol (Level 200): Building MCP servers & clients, stdio/SSE transports, resources, tools, and prompts.
• 🎓 Claude Code in Action (Level 200): CLI workflows, project memory (CLAUDE.md), hooks, and custom subagent orchestration.

🔗 Access the official portal: Anthropic Partner Skilljar Academy

2. High-Yield Practice Questions & Mock Exams

To test your readiness under realistic time and scenario constraints, practice with high-yield mock exams that reflect the latest Anthropic blueprints:

• 🌟 Recommended Practice Partner: CertsClub Claude Certified Associate / Architect Foundations Exam Preparation
  • Realistic 60-question timed mock exams
  • In-depth answer rationales for all multiple-choice and multi-select questions
  • Scenario-based questions matching Pearson VUE / Online proctored exam difficulty

💡 Quick Architectural Cheatsheet

1. Model Family Selection Decision Matrix

┌─────────────────┬──────────────────────┬────────────────────────┬─────────────────────────┐
│ Model           │ Primary Use Cases    │ Strengths              │ Ideal Architecture Role │
├─────────────────┼──────────────────────┼────────────────────────┼─────────────────────────┤
│ Claude 3.5      │ Complex Reasoning,   │ Industry-leading       │ Primary Agent Brain,    │
│ Sonnet          │ Coding, Architecture,│ coding & vision,       │ Orchestrator, Deep      │
│                 │ Multi-Step Workflows │ fast token generation  │ Analysis & Tool Caller  │
├─────────────────┼──────────────────────┼────────────────────────┼─────────────────────────┤
│ Claude 3.5      │ High Volume, Low     │ Blazing fast speed,    │ Intent Classification,  │
│ Haiku           │ Latency, Cost Tuning,│ sub-second TTFT,       │ Routing, Filtering,     │
│                 │ Light Tool Calls     │ ultra-low token cost   │ Real-time Chatbots      │
├─────────────────┼──────────────────────┼────────────────────────┼─────────────────────────┤
│ Claude 3        │ Massive Synthesis,   │ Nuanced writing, deep  │ Specialized Executive   │
│ Opus            │ Highly Ambiguous     │ philosophical synthesis│ Summaries, Complex Math │
│                 │ Long-Form Context    │ across massive corpora │ Evaluation Judgments    │
└─────────────────┴──────────────────────┴────────────────────────┴─────────────────────────┘

2. Model Context Protocol (MCP) Core Anatomy

┌────────────────────────────────────────────────────────┐
│                      MCP HOST                          │
│  (e.g., Claude Code, Claude Desktop, Custom AI Server) │
│                                                        │
│   ┌────────────────────────────────────────────────┐   │
│   │                   MCP CLIENT                   │   │
│   └───────────────────────┬────────────────────────┘   │
└───────────────────────────┼────────────────────────────┘
                            │ JSON-RPC 2.0 (stdio / SSE)
    ┌───────────────────────┴────────────────────────┐
    ▼                                                ▼
┌─────────────────────────┐            ┌─────────────────────────┐
│     MCP SERVER A        │            │     MCP SERVER B        │
│  (Local Files / Git)    │            │  (Postgres DB / REST)   │
├─────────────────────────┤            ├─────────────────────────┤
│ • Tools (Execute)       │            │ • Tools (Execute)       │
│ • Resources (Read)      │            │ • Resources (Read)      │
│ • Prompts (Templates)   │            │ • Prompts (Templates)   │
└─────────────────────────┘            └─────────────────────────┘

3. Prompt Caching Economics & Latency Optimization

• Minimum Cacheable Threshold: 1,024 tokens (Claude 3.5 Sonnet / Opus) or 2,048 tokens (Claude 3.5 Haiku).
• TTL (Time to Live): 5-minute rolling window refreshed on cache hits.
• Pricing Multipliers:
  • Cache Write (1.25x base input cost)
  • Cache Read Hit (0.10x base input cost — 90% discount!)
• Breakpoint Placement: Place cache_control: {"type": "ephemeral"} at static boundaries (System Prompts, Large Tool Definitions, RAG Documentation blocks).

🤝 Community & Contributions

Contributions, typo fixes, and additional scenario additions are welcome! Open a pull request or issue following the standard GitHub workflow.

📄 License

This repository is open-sourced under the MIT License. All trademarks and brand names belong to Anthropic PBC.