The complete preparation pathway for the Anthropic Claude Certified Architect – Foundations exam. Scenario-based questions, domain-by-domain study, and timed mock tests included.
Claude Certified Architect: Exam Overview
What the exam tests, five domain weightings, six scenario frames, scoring (720/1000 to pass), and how to prepare.
Agentic Loop Implementation: stop_reason, tool_use, and end_turn
stop_reason control flow (tool_use vs end_turn), conversation history construction, and programmatic vs prompt-based enforcement — Domain 1 deep dive.
How to Build Your First Agent Loop
Implement stop_reason-based control flow, tool result handling, and correct loop termination — the core of Domain 1.
What Is Loop Engineering?
The paradigm shift from prompt engineering to loop design — why the execution scaffold matters as much as the model.
Multi-Agent Orchestration Patterns
Coordinator-subagent hub-and-spoke, parallel fan-out, iterative refinement, and why subagents don't inherit parent context.
Claude Code Subagents: Multi-Agent Workflows
Spawning subagents via the Task tool, allowedTools configuration, parallel execution, and explicit context passing.
Claude Code Hooks: Automate Actions on Tool Calls
PostToolUse hooks for data normalization, tool call interception for compliance enforcement, and when hooks beat prompt instructions.
What Is MCP? Model Context Protocol Explained
MCP architecture, tools vs resources, isError flag, structured error responses, and .mcp.json configuration.
Build Your First MCP Server
Working MCP server with typed tool schemas, error categories (transient/validation/permission), and isRetryable metadata.
MCP Tool Descriptions: How to Write Them for Reliable Agent Selection
Four-question minimum for viable descriptions, naming conventions to prevent overlap, tool_choice options, and structured error responses with isError/isRetryable.
Tool Definition and Schema Design
Writing tool descriptions that prevent selection confusion, splitting overlapping tools, and scoping tools per agent role.
Claude Code MCP Servers: Connect Any Tool
Project-scoped .mcp.json vs user-scoped ~/.claude.json, environment variable expansion, and MCP resources for content catalogs.
What Is CLAUDE.md? Persistent Memory for Claude Code
User/project/directory hierarchy, @import syntax, .claude/rules/ with glob patterns, and diagnosing configuration issues.
Claude Code Plan Mode: Complete Guide
When to use plan mode (multi-file refactors, architectural decisions) vs direct execution (single-file bug fixes).
Steering Claude Code: CLAUDE.md, Skills, Hooks, Subagents, Rules
Project-scoped vs user-scoped commands, context: fork for skill isolation, allowed-tools, and argument-hint frontmatter.
Claude Code Commands: Complete Reference
The -p/--print flag for CI mode, --output-format json, --json-schema, and session management (--resume, fork_session).
Claude Code in CI/CD Pipelines: The -p Flag, JSON Output, and Automated Review
--print for non-interactive mode, --output-format json with --json-schema, CLAUDE.md in CI, multi-pass review, and Batch API vs real-time cost tradeoffs.
Zero-Shot, Few-Shot & Chain-of-Thought Prompting
Few-shot examples for ambiguous scenarios, format consistency, false positive reduction, and generalizing to novel patterns.
Structured Output with tool_use and JSON Schemas: The Definitive Guide
Why tool_use beats JSON-in-prompt, nullable field design, enum+detail patterns, validation-retry loops, and when retries should stop.
Structured Output & JSON Mode
tool_use with JSON schemas, tool_choice (auto/any/forced), nullable fields to prevent hallucination, and validation-retry loops.
Master Prompt Engineering with Claude
Explicit criteria over vague instructions, multi-pass review architectures, and independent review instances for quality.
How to Evaluate Prompt Quality
Build eval sets, measure false positive rates by category, and design batch processing strategies (Message Batches API).
What Is Context Engineering?
Progressive summarization risks, lost-in-the-middle effects, structured fact extraction, and position-aware input ordering.
Claude Code Context Window: Managing Limits
Trimming verbose tool outputs, scratchpad files for long sessions, /compact, and subagent delegation for context isolation.
Error Propagation in Multi-Agent Systems: Structured Context Over Generic Failures
Anti-patterns (empty-as-success, workflow termination, generic strings), structured error context fields, four error categories, and coverage gap annotations.
Conversation History Management
Structured error propagation across multi-agent systems, escalation triggers, and information provenance in multi-source synthesis.
Practice exam
3 timed mock exams with shuffled questions, instant scoring, and per-question explanations. Pass score: 720/1000. The fastest way to find your weak domains before exam day.
It's Anthropic's official practitioner certification, administered through Anthropic Academy. The 120-minute exam has 60 scenario-based multiple-choice questions across five domains: Agentic Architecture & Orchestration (27%), Tool Design & MCP Integration (18%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%), and Context Management & Reliability (15%). The minimum passing score is 720 out of 1,000.
Solution architects and developers who design and build production applications with Claude. The target candidate has 6+ months of hands-on experience with the Claude API, Agent SDK, Claude Code, and MCP — and wants external validation of their architectural judgment, not just API familiarity.
Questions are scenario-based, not definitional. Every question frames a realistic production situation — a customer support agent that escalates incorrectly, a CI pipeline that hangs, a multi-agent system with coverage gaps — and asks you to diagnose the root cause and choose the most effective fix. Knowing what tools exist isn't enough; you need to reason about tradeoffs.
17 articles across all five exam domains, approximately 10 hours of study. The pathway is structured to mirror the exam's domain weighting: heaviest on Agentic Architecture (Domain 1) and lightest on Context Management (Domain 5). Practice quiz questions appear throughout to test recall in exam format.
The pathway includes scenario-based quiz questions drawn from the official exam guide throughout each article. After completing the pathway, use the Claude Certified Architect mock tests at /tests/claude-certified-architect — timed, full-length practice exams with shuffled questions and per-answer explanations. Mock tests are included with an active subscription.
You should be comfortable reading and writing TypeScript or Python, understand what APIs and JSON are, and have at least some hands-on experience with Claude. The Claude Code Mastery, Building AI Agents, and MCP: Model Context Protocol pathways on this platform cover the prerequisite knowledge if you need to build up before diving into this pathway.
Understand what AI actually is — tokens, transformers, agents, and the landscape. Start here if you're new.
11 articles · ~4h →Go from vague requests to precise, reproducible AI outputs. The skill that underpins everything.
12 articles · ~5h →Go from zero to productive with Claude Code — the terminal AI coding agent that ships real projects.
13 articles · ~7h →Understand and build the loops, harnesses, and protocols that make AI agents reliable and autonomous.
14 articles · ~6h →Practical AI adoption for your specific function — marketing, engineering, HR, finance, and more.
10 articles · ~4h →Navigate the crowded model market — Claude, GPT, Gemini, open-source — and understand the tradeoffs.
10 articles · ~6h →
