CCAR-F
Claude Certified Architect (Foundations)
Who it is for: Solution architects working with Claude Code, the Agent SDK, the API and MCP.
60 questions · 120 minutes
31 lessons · 5 domains
Take a practice examHow do I take the exam?
- Your company must be part of the Claude Partner Network. Joining is free.
- On the Anthropic Partner Academy, open the certification page, read the guide and the terms, and register. The list price is USD 125; checkout shows the final price, with any discount for your company's partner tier.
- After confirming, create your Pearson VUE account and pick a date: online proctored or at a Pearson test centre.
- You can cancel or reschedule up to 24 hours before. After that, the fee is forfeited.
- The credential lasts 12 months. Renewing on time is free: a short, non-proctored assessment on what has changed.
Domains
Agentic Architecture 27%
- Rules that must always hold belong in code, not in the prompt
- Coordinator and subagents
- What a subagent receives and what it returns
- Task decomposition, fixed or dynamic
- Sessions, resume, fork or start fresh
- The agentic loop: keep going while stop_reason is tool_use
- Agent SDK hooks: intercepting calls and normalizing results
- Multi-concern requests and handoff to a human
Tool Design & MCP 18%
- The description decides which tool the model picks
- Tool errors the agent can act on
- Few tools per agent, and tool_choice
- MCP servers in Claude Code and in the agent: scope, credentials and resources
- Read, Write, Edit, Bash, Grep and Glob: which one for what
Claude Code 20%
- CLAUDE.md: hierarchy, scope and imports
- Commands and skills: where they go and how to configure them
- Rules that load depending on the file
- Plan mode or direct execution
- Refining step by step: examples, tests and the interview
- Claude Code in CI/CD
Prompt Engineering 20%
- Explicit criteria instead of "be conservative"
- Few-shot examples for ambiguous cases and consistent format
- Structured output with tool_use and JSON Schema
- Validate, retry with the error and know when not to retry
- Message Batches: which jobs go in a batch
- Review with independent instances and multiple passes
Context Management 15%
- The context window and how to manage it
- When to escalate to a human and how to resolve ambiguity
- Errors that travel well between agents
- Exploring a large codebase without losing the thread
- Human review and calibrated confidence
- Information provenance and conflicting sources