Claude Architect
Claude Certified Architect: Foundations
An independent study guide for the CCAR-F exam blueprint: agentic architecture, tool and MCP design, Claude Code configuration, prompt engineering for structured output, and context management. Written from scratch and not affiliated with, endorsed by or sponsored by Anthropic. It reproduces no exam questions.
6 modules · 8 hr 54 min of reading
4 of 34 lessons are open to read without an account, and a free account opens the other 30. Creating one costs nothing: there is no paid plan, no tier and nothing to buy.
- lessons
- 34
- flashcards
- 165
- quiz questions
- 132
- practice questions
- 420
Practice exam
A sitting is 60 questions in 2 hr, reported on the 100 to 1000 scale the programme publishes, with 720 to pass.
- Drawn fresh from a pool of 420, allocated across the domains by the weights below
- Worked through unseen questions first, so a second sitting is a different paper
- Feedback held back until you submit, then every option explained and the score split by domain
- Agentic Architecture and Orchestration
- 27 per cent of the paper
- Tool Design and MCP Integration
- 18 per cent of the paper
- Claude Code Configuration and Workflows
- 20 per cent of the paper
- Prompt Engineering and Structured Output
- 20 per cent of the paper
- Context Management and Reliability
- 15 per cent of the paper
The weights are the published blueprint's. The questions are not: all 420 were written here from the published objectives, and none of them is a real exam question.
Course outline
Module 1
Open to readOrientation: what CCAR-F measures
What the Claude Certified Architect Foundations exam is, the blueprint it is built on, the six production scenarios it frames questions inside, and how to prepare. Read this first: it also sets out what this guide is and is not.
4 lessons · 54 min
Before you start
- What the CCAR-F exam measures
- The six production scenarios
- How to prepare by building
- What this guide is sure about
Module 2
Free accountDomain 1: Agentic architecture and orchestration
The largest domain at 27%. Agentic loops driven by stop_reason, coordinator and subagent systems, context passing, programmatic enforcement, hooks, task decomposition, and session lifecycle.
7 lessons · 1 hr 53 min
Loops and orchestration
- The agentic loop
- Coordinators and subagents
- Invoking subagents and passing context
Control and decomposition
- Enforcement and handoffs
- Hooks for interception and normalisation
- Task decomposition strategies
- Sessions: resume and fork
Module 3
Free accountDomain 2: Tool design and MCP integration
18% of the exam. Tool descriptions as the selection mechanism, structured error responses, distributing tools across agents, tool_choice, MCP server scoping, and the built-in tools.
5 lessons · 1 hr 18 min
Designing tool interfaces
- Writing tool descriptions that make selection reliable
- Structured error responses for MCP tools
- Distributing tools across agents and configuring tool choice
MCP and built-in tools
- Integrating MCP servers at the right scope
- Selecting and applying the built-in tools
Module 4
Free accountDomain 3: Claude Code configuration and workflows
20% of the exam. The CLAUDE.md hierarchy, slash commands and skills, path-scoped rules, plan mode versus direct execution, iterative refinement, and non-interactive use in CI.
6 lessons · 1 hr 36 min
Configuring the workspace
- The CLAUDE.md hierarchy and what teammates actually get
- Skills and slash commands as shared procedure
- Path-scoped rules for conventions that follow a file type
Working and shipping
- Plan mode versus direct execution
- Iterative refinement that converges
- Claude Code in a pipeline
Module 5
Free accountDomain 4: Prompt engineering and structured output
20% of the exam. Explicit criteria over vague instruction, few-shot examples, tool use with JSON schemas for guaranteed structure, validation and retry, batch processing, and multi-pass review.
6 lessons · 1 hr 36 min
Getting the output you asked for
- Explicit criteria, not quality adjectives
- Few-shot examples for the cases rules cannot settle
Structure, validation and scale
- Structured output with tools and JSON schemas
- Validation, retry and knowing when retrying cannot help
- Batch processing and the latency it costs you
- Multi-instance and multi-pass review
Module 6
Free accountDomain 5: Context management and reliability
15% of the exam. Preserving facts across long conversations, escalation and ambiguity, error propagation between agents, exploring large codebases, human review and confidence calibration, and provenance in synthesis.
6 lessons · 1 hr 37 min
Holding on to what matters
- Keeping the facts a long conversation depends on
- Exploring a large codebase without losing the map
Failing well
- Escalation triggers that hold up, and two that do not
- Errors a coordinator can actually act on
- Designing review around calibrated confidence
- Provenance that survives synthesis
Where this course comes from
Independent study guide. Structure follows the publicly published CCAR-F exam blueprint; all explanations, examples and practice questions are our own.
Start with the orientation module
It reads with no account at all, and it sets out what the exam measures, how to prepare for it, and what this guide is and is not sure about.
4 of 34 lessons are open to read without an account, and a free account opens the other 30. There is no paid plan, no tier and nothing to buy.
