Claude · Claude Certified Architect, Foundations
Preparing for CCAR-F
The hands-on, scenario-based entry credential in the Claude Certification Program. It validates sound tradeoff decisions when building production systems with Claude Code, the Claude Agent SDK, the Claude API, and MCP.
Find your weak domains first
Sit the 60-question mock under exam conditions, or study all 360 questions with explanations and doc links.
Unofficial. Not real exam questions and not affiliated with Anthropic.
Exam at a glance
- Credential
- Claude Certified Architect, Foundations
- Exam code
- CCAR-F
- Items
- 60 (multiple-choice and multiple-response)
- Structure
- 4 scenarios drawn at random from a bank of 6
- Time
- 120 minutes
- Passing score
- 720 on a 100-1,000 scale
- Delivery
- Proctored (online or test center)
- Fee
- $125 USD
- Validity
- 12 months
Confirm against the official CCAR-F page.
The 5 content domains
Agentic Architecture & Orchestration
Agentic loops (stop_reason), coordinator/subagent patterns, the Task tool, hooks, task decomposition, session resume/fork
Tool Design & MCP Integration
Tool descriptions, structured errors (isError), tool distribution, tool_choice, .mcp.json scoping, built-in tools
Claude Code Configuration & Workflows
CLAUDE.md hierarchy & @import, slash commands, .claude/rules/ globs, plan vs. direct execution, CI (-p, JSON output)
Prompt Engineering & Structured Output
Explicit criteria, few-shot, tool_use + JSON schemas, nullable fields, retry-with-feedback, Batches API, multi-pass review
Context Management & Reliability
Summarization risk, case-facts blocks, escalation triggers, error propagation, provenance, confidence calibration
The 6 scenarios (4 appear on exam day)
- 1Customer Support Resolution Agent, Agent SDK + MCP tools; 80%+ first-contact resolution.
- 2Code Generation with Claude Code, slash commands, CLAUDE.md, plan vs. direct execution.
- 3Multi-Agent Research System, coordinator + search/analyze/synthesize/report subagents.
- 4Developer Productivity with Claude, Agent SDK, built-in tools, MCP servers.
- 5Claude Code for Continuous Integration, CI/CD reviews, test generation, PR feedback.
- 6Structured Data Extraction, tool_use + JSON schemas, validation, edge cases.
The mental model that passes
R1 · Root cause, not symptom
Fix what is actually broken, not the surface symptom.
R2 · Deterministic beats probabilistic
When compliance matters, prefer hooks, prerequisite gates, tool_choice, and JSON schemas over prose instructions.
R3 · Proportionate wins
The lowest-effort, highest-leverage fix is usually correct; ML classifiers and heavy infra "first steps" are traps.
How to prepare, a 3-step path
- Diagnose. Take the mock exam to find your weakest domains, then read the blueprint and self-assess against each domain’s task statements.
- Study interactively. Create a Claude Project, upload the exam guide PDF, and use the prompt packs or the ccar-f-examprep-coach skill.
- Build three projects. An agentic loop with escalation logic; a Claude Code team setup (CLAUDE.md hierarchy,
.claude/rules/, acontext: forkskill, an MCP server); and a structured-extraction pipeline (tool_use+ JSON schema + validation-retry).
Facts come from the official Exam Guide (v1.0, effective July 2026). Always confirm against the current official page, the guide is subject to change.