SAI

SappersAI

Claude Certified Architect – Foundations Certification Exam Guide

Atul Divekar

The Claude Certified Architect – Foundations certification is the official credential from Anthropic that proves you can build professional, production-grade AI systems. It focuses on four core technologies: Claude Code, the Claude Agent SDK, the Claude API, and the Model Context Protocol (MCP).

📋 Exam Fast Facts

  • Format: Multiple Choice (60 questions).
  • Time Limit: 120 minutes.
  • Passing Score: 720 / 1,000.
  • Experience Level: Recommended 6+ months of hands-on experience with Claude APIs.
  • Core Focus: Making smart architectural tradeoffs in real-world scenarios.

🎯 The 5 Knowledge Domains

The exam is broken down into these specific areas. Mastering these is the key to passing.

1. Agentic Architecture & Orchestration (27%)

  • Agentic Loops: Knowing how to handle stop_reason (knowing when the agent is done vs. when it needs to use a tool).
  • Multi-Agent Systems: Using a “Coordinator-Subagent” pattern where one agent manages others.
  • Context Passing: Ensuring subagents get the specific info they need, as they don’t automatically “know” what the parent agent knows.

2. Tool Design & MCP Integration (18%)

  • Writing Descriptions: Your tool descriptions are how the AI “decides” which tool to pick. Poor descriptions lead to errors.
  • Error Handling: Using the isError flag and providing structured feedback so the agent can try a different approach if a tool fails.
  • MCP Servers: Managing project-level (.mcp.json) vs. user-level (~/.claude.json) tools.

3. Claude Code Configuration & Workflows (20%)

  • CLAUDE.md: Setting up instructions that tell Claude how to behave in specific projects.
  • Plan Mode vs. Direct Execution: Using “Plan Mode” for big architectural changes and “Direct Execution” for quick, 1-file fixes.
  • Custom Skills: Creating isolated “forked” environments for complex tasks to keep the main conversation clean.

4. Prompt Engineering & Structured Output (20%)

  • Few-Shot Prompting: Giving 2-4 examples of exactly how you want the output to look.
  • JSON Schemas: Forcing Claude to return data in a strict format so your code can read it.
  • Batch Processing: Saving 50% on costs by using the Message Batches API for non-urgent tasks.

5. Context Management & Reliability (15%)

  • The “Lost in the Middle” Effect: Realizing AI is best at remembering the very beginning and very end of a prompt.
  • Escalation: Designing clear rules for when the AI should stop and ask a human for help.
  • Provenance: Ensuring the AI can cite exactly which document or source a piece of information came from.

🛠️ Expert Preparation: 3 Essential Exercises

To pass, you must go beyond theory. Try these hands-on tasks:

  1. Build a Multi-Agent System: Create a “Coordinator” agent that uses the Task tool to spawn a “Researcher” subagent.
  2. Configure a Workflow: Set up a CLAUDE.md file and create a custom slash command in the .claude/commands/ folder.
  3. Create an Extraction Pipeline: Build a tool that takes a messy PDF and returns a clean, validated JSON object.

🚀 Pro-Tip for the Exam

Watch out for “Scenario” questions. The exam uses 4 random scenarios (like a Customer Support Agent or a Research System). Always ask yourself: “Does this require a human-in-the-loop, or can the AI handle this autonomously based on the company policy?”.

Leave a comment