Claude Certified Architect: Foundations

Who it is for

Solution architects and engineers who design and build production applications with Claude, hands-on with the Agent SDK, Claude Code, MCP and the API. The guide suggests six months or more of building.

How the exam is built

The exam guide publishes six scenarios. You get four of them, with about 15 questions on each.

1. A customer support agent that resolves cases and escalates to a person.
2. Code generation with Claude Code.
3. A multi-agent research system with a coordinator and subagents.
4. Developer productivity with built-in tools and MCP.
5. Claude Code in continuous integration.
6. Structured data extraction.

One published review, from someone who passed, says the scenarios matter less than they look. The questions under them test specific concepts. So my notes are organised by concept and domain, not by scenario.

Two kinds of question

The same review describes the exam as two halves.

• The reasoning half: tool design, orchestration and error handling. With good engineering instincts you can work out the answer.
• The memory half: which file lives where, which flag does what, what each stop_reason means. You cannot derive these, and the exam is closed book.

Exam tip: Good instincts will not tell you which of four flags is the real one. Every note I write ends with a block of exact facts for this half.

What the guide puts in scope

The Claude Agent SDK, MCP, Claude Code, the Claude API, the Message Batches API, JSON Schema and Pydantic, CLAUDE.md, and the built-in tools.

The guide also says what it leaves out: fine-tuning, API authentication and billing, MCP hosting, embeddings and vector databases, computer use, vision, streaming, rate limits and pricing, and how prompt caching is implemented.

The task statements, by domain

These are the task statements from the exam guide, in my own shortened wording.

Domain 1. Agentic architecture and orchestration, 27%

1. Agentic loops
2. Coordinator and subagent orchestration, hub and spoke
3. Subagent invocation, context passing and spawning
4. Multi-step workflows with enforcement and handoff
5. Agent SDK hooks for interception and normalisation
6. Task decomposition
7. Session state, resumption and forking

Domain 2. Tool design and MCP integration, 18%

1. Tool interfaces with clear descriptions and boundaries
2. Structured error responses
3. Tool distribution across agents, and tool_choice
4. MCP server integration
5. Built-in tool selection

Domain 3. Claude Code configuration and workflows, 20%

1. The CLAUDE.md hierarchy and modular organisation
2. Slash commands and skills
3. Path-specific rules
4. Plan mode against direct execution
5. Iterative refinement
6. CI/CD integration

Domain 4. Prompt engineering and structured output, 20%

1. Explicit criteria to cut false positives
2. Few-shot prompting
3. Structured output through tool use and JSON schemas
4. Validation, retry and feedback loops
5. Batch processing
6. Multi-instance and multi-pass review

Domain 5. Context management and reliability, 15%

1. Preserving critical information in long conversations
2. Escalation and resolving ambiguity
3. Error propagation across agents
4. Context in large codebase exploration
5. Human review and confidence calibration
6. Provenance and uncertainty when combining sources

How I am studying

Four phases, with the heaviest domains first: Domains 1 and 2, then Domain 3, then Domains 4 and 5, then a pass over everything. I take a practice set by domain after the second phase, not cold at the start.

The full plan, with the 21 notes I intend to write, is in Half reasoning, half memory.

Sources

• CCAR-F exam guide, Anthropic: the format, the scenarios, the domains, the task statements and what is in scope.
• “The Claude certification exams: an honest review” by Matthew Purcell, 22 July 2026, for the two halves and the point about scenarios: https://www.linkedin.com/pulse/claude-certification-exams-honest-review-matthew-purcell-byo2e/
• The study method and the notes are mine.