Claude Certified Architect Practice Tests

1. Exam Overview

• Format: Multiple choice (100–1,000 scaled score; 720 to pass).

• Experience Level: Recommended 6+ months of hands-on experience.

• Key Technologies: Claude API, Claude Agent SDK, Claude Code, and Model Context Protocol (MCP).

2. Core Domains & Weighting

The exam is divided into five functional domains:

DomainWeightKey Focus Area1. Agentic Architecture27%Loops, multi-agent orchestration, and hooks.2. Tool Design & MCP18%Tool definitions, error handling, and MCP servers.3. Claude Code20%CLAUDE md, slash commands, and CI/CD integration.4. Prompt Engineering20%Structured output (JSON), few-shotting, and precision.5. Context & Reliability15%Context window management and escalation logic.

3. High-Value Architectural Concepts

To succeed, you must understand these specific technical patterns:

Agentic Loops & Multi-Agent Orchestration

The "Hub-and-Spoke" model is the standard for complex systems.

• The Loop: You must inspect stop_reason. If it is tool_use, you execute the tool and feed the result back; if end_turn, the task is finished.

• Context Isolation: Subagents do not inherit the coordinator's history. You must explicitly pass relevant data in the subagent's prompt.

• Parallelism: To speed up workflows, a coordinator can emit multiple Task tool calls in a single turn.

Tool Design & MCP

• Descriptions are Everything: The model chooses tools based on the text description. Ambiguous descriptions lead to "hallucinated" tool calls or misrouting.

• MCP Resources vs. Tools: Use Resources for data catalogs (read-only schemas or docs) and Tools for actions (searching, writing).

• Error Handling: Use the isError flag and provide structured metadata (e.g., isRetryable: true) so the agent knows whether to try again or escalate.

Claude Code & CLAUDE md

• The Hierarchy: * User-level: ~/.claude/CLAUDE md (Personal settings, NOT shared).

  • Project-level: Root CLAUDE md (Shared via Git, contains team standards).

• Plan Mode vs. Direct Execution: Use Plan Mode for architectural changes or multi-file refactors. Use Direct Execution for small, well-defined bug fixes.

• CI/CD: Use the -p (print) flag for non-interactive runs to prevent the pipeline from hanging on user prompts.

4. Key Study Tips for Scenario Questions

The exam will randomly select 4 out of 6 possible scenarios (Support Agent, Code Gen, Research System, Dev Productivity, CI/CD, and Data Extraction).

Pro-Tip: Focus on the Mental Models. For example, always choose programmatic enforcement (hooks/gates) over "prompt instructions" for critical safety or financial rules (like a $500 refund limit), because prompts are probabilistic, while code is deterministic.