Exam Details at a Glance

Source: Official Anthropic Exam Guide, Version 1.0, Effective July 2026. Exam code: CCAR-F.

Item Value
Credential Claude Certified Architect – Foundations
Exam code CCAR-F
Number of items 60
Time limit 120 minutes
Item format Multiple-choice and multiple-response items; each item states how many responses to select
Exam structure 4 scenarios drawn from a bank of 6
Content domains 5 (weighted)
Delivery Proctored: online proctored and/or test center, per program policy
Exam fee $125 USD
Scoring Scaled 100–1,000; passing = 720
Validity 12 months from the date the credential is awarded
Result reporting Pass/fail with scaled score, plus percent-correct by domain on the score report (informational only — not used to determine pass/fail)

Every question must be answered before advancing (no skip-and-return blank state at time of submit — check pacing per scenario, ~30 min per scenario/15 questions if it splits evenly across the 4 scenarios shown).

Note on item format: the guide's own wording allows for both single-answer and multi-select ("choose N") items, though none of the 12 official sample questions in the guide demonstrate a multi-select item — all 12 remain standard single-answer, 4-option format.

Target Candidate

A solution architect with 6+ months hands-on experience who can:
- Build agentic apps with the Agent SDK: multi-agent orchestration, subagent delegation, tool integration, lifecycle hooks
- Configure Claude Code for teams: CLAUDE.md, Agent Skills, MCP integrations, plan mode
- Design MCP tool/resource interfaces for backend integration
- Engineer prompts for reliable structured output: JSON schemas, few-shot, extraction patterns
- Manage context windows across long docs, multi-turn conversations, multi-agent handoffs
- Integrate Claude into CI/CD for code review, test generation, PR feedback
- Make escalation/reliability calls: error handling, human-in-the-loop, self-evaluation

Content Domains & Weights

  1. Agentic Architecture & Orchestration — 27%
  2. Tool Design & MCP Integration — 18%
  3. Claude Code Configuration & Workflows — 20%
  4. Prompt Engineering & Structured Output — 20%
  5. Context Management & Reliability — 15%

Domain 1 is the single biggest chunk — prioritize agentic loops, coordinator/subagent patterns, hooks, and task decomposition.

The Exam's Recurring Judgment Pattern

Across every sample question, the "correct" answer follows a consistent philosophy worth internalizing:

  1. Prefer the proportionate root-cause fix over infrastructure. Don't reach for a classifier, routing layer, or new model tier when a better tool description or explicit criteria would fix it.
  2. Use deterministic enforcement (hooks, programmatic prerequisites) only when the cost of failure is high (money, compliance, identity) — otherwise prompt/few-shot fixes are the "first step."
  3. Self-reported confidence/sentiment from the model is NOT a reliable control signal — don't pick answers that lean on the model grading itself for escalation or routing.
  4. Don't blame a downstream component when the log evidence points upstream (e.g., coordinator decomposition, not the subagent execution).
  5. Match architecture to actual constraints, e.g., batch API for latency-tolerant/non-blocking work only.
  6. Least privilege for tools — scope tools tightly per agent; only add cross-role tools for high-frequency simple cases.