CCAR-F Claude Certified Architect — Foundations

Cheatsheet 1 — Blueprint & Exam Mechanics

One page. Read it the morning of the exam.


The exam

Credential Claude Certified Architect – Foundations
Code CCAR-F · Version 1.0 · effective July 2026
Items 60
Format Multiple-choice and multiple-response — each item states how many to select
Structure 4 scenarios per sitting, drawn at random from a bank of 6
Time 120 minutes = 2.0 min/item
Delivery Proctored: online proctored and/or Pearson VUE test centre
Pass 720 scaled, on a 100–1,000 scale
Scoring Criterion-referenced — fixed standard from a standard-setting study. Not a curve.
Fee $125 USD per attempt
Validity 12 months
Report Pass/fail + scaled score + percent correct by domain (domain figures are informational only)
Retakes 14 days after 1st fail · 30 after 2nd · 90 after 3rd · max 4 attempts per rolling 12 months
Renewal Free non-proctored assessment on Anthropic Partner Academy if renewed on time; full re-exam if lapsed
Assumed background Solution architect with 6+ months hands-on

Technologies tested: Claude Code · Claude Agent SDK · Claude API · Model Context Protocol (MCP).


The blueprint

# Domain Weight ≈ items
1 Agentic Architecture & Orchestration 27% ~16
2 Tool Design & MCP Integration 18% ~11
3 Claude Code Configuration & Workflows 20% ~12
4 Prompt Engineering & Structured Output 20% ~12
5 Context Management & Reliability 15% ~9

Domains 1 + 3 + 4 = 67% of the exam.


The six scenarios

# Scenario Domains Signature content
1 Customer Support Resolution Agent 1, 2, 5 Agent SDK; get_customer, lookup_order, process_refund, escalate_to_human; 80%+ first-contact resolution
2 Code Generation with Claude Code 3, 5 Custom slash commands, CLAUDE.md, plan mode vs direct execution
3 Multi-Agent Research System 1, 2, 5 Coordinator + web-search / document-analysis / synthesis / report-generation subagents; cited reports
4 Developer Productivity with Claude 2, 3, 1 Agent SDK; built-ins (Read, Write, Bash, Grep, Glob) + MCP servers
5 Claude Code for Continuous Integration 3, 4 Automated review, test generation, PR feedback; minimize false positives
6 Structured Data Extraction 4, 5 JSON-schema validation, high accuracy, graceful edge cases, downstream integration

You will see 4 of these 6. All six are worth preparing — see Chapter 13.


Out of scope — do not study, do not pick

  • Fine-tuning or training custom models
  • Claude API authentication, billing, account management
  • Language/framework implementation detail beyond tool & schema configuration
  • Deploying or hosting MCP servers
  • Claude's internal architecture, training, or weights
  • Constitutional AI, RLHF, safety training
  • Embedding models, vector databases — there is no RAG domain
  • Computer use · vision / image analysis
  • Streaming API / SSE
  • Rate limiting, quotas, pricing calculations
  • OAuth, API key rotation, auth protocols
  • Cloud-provider-specific configuration (AWS, GCP, Azure)
  • Performance benchmarking, model comparison / model selection by price
  • Prompt-caching implementation details
  • Token counting / tokenization

🎯 An option that lives in this list is almost always a distractor.


Time budget

Checkpoint Elapsed Items done
Scenario 1 finished 30 min 15
Scenario 2 finished 60 min 30
Scenario 3 finished 90 min 45
Scenario 4 finished 110 min 60
Review flagged items 110–120 min

Rule: 90 seconds on the first read. If you haven't converged, flag it and move. Answer every item — there is no penalty for a wrong answer.

Read the scenario setup once, carefully, then work its items. Re-reading the setup for every item burns your buffer.


Target score

Anthropic does not publish the raw→scaled mapping. Working targets on a 60-item test:

Raw % Read
44/60 73% Approximate floor for 720
48/60 80% Comfortable margin
52/60 87% Safe

Per-domain minimums to aim for: D1 12/16 · D2 8/11 · D3 9/12 · D4 9/12 · D5 7/9.