A comprehensive study guide for developers and solution architects preparing for the Claude Certified Architect – Foundations certification exam.

This guide is designed for developers and solution architects who are new to the LLM space and want to build a solid foundation for the certification. If you have experience building LLM applications, it also serves as a focused review of the nuances the exam emphasizes.

The guide covers seven interconnected knowledge areas:

1. Designing Tool Interfaces for LLM Agents — parameter design, structured output, tool composition, confirmation flows
2. Error Handling in Agent Tools — transient vs. permanent errors, structured error responses, uncertain state, MCP error patterns
3. Conversation Context Management — stateless API fundamentals, progressive summarization, state objects, stale data handling
4. System Prompt Engineering — structure, principles vs. conditionals, dilution, few-shot examples
5. Model Context Protocol (MCP) — architecture, trust model, tool descriptions, multi-server availability
6. Agentic Patterns & Task Decomposition — prompt chaining, routing, orchestrator-workers, dynamic decomposition
7. Agentic Workflow Design — escalation, compliance enforcement, graceful degradation, handoff design

Each section includes:

• What to Know — core concepts explained as teaching material
• Key Relationships — how the area connects to others
• Common Pitfalls — misconceptions and subtle distinctions
• Go Deeper — sub-topics that deserve extra attention

Plus a Study Strategy, Quick Reference Cheat Sheet, and Recommended Reading & Resources with links to official documentation.

exam-preparation-guide.md

This is an independent, community-created study guide. It is not affiliated with, endorsed by, or sponsored by Anthropic. No exam questions are included, paraphrased, or hinted at — this guide teaches the underlying knowledge domains only.

The guide was authored with assistance from Claude.

Found an error? Have a suggestion? Feel free to open an issue or submit a pull request.

This work is licensed under CC BY 4.0. You are free to share and adapt it with attribution.