Study guide
1.1 Certification Context
Claude Certified Architect - Foundations Study Guide Exam
Track your progress as you read
Finished sections get a check mark.
1 Chapter 1: Exam and Ecosystem Overview
1.1 Certification Context 1.2 Study Map
2 Chapter 2: Agentic Architecture and Orchestration
2.1 Workflows vs Agents 2.2 Agentic Loops and the Claude Agent SDK 2.3 Multi-Agent Orchestration and Session Management
3 Chapter 3: Claude Code Configuration and Workflows
3.1 Claude Code Workflows
4 Chapter 4: Tool Design and MCP Integration
4.1 Tool Design and MCP Integration
5 Chapter 5: Prompt Engineering and Structured Output
5.1 Prompt Engineering 5.2 Structured Output and Evaluation
6 Chapter 6: Context Management and Reliability
6.1 Context, Caching, and Compaction 6.2 Model Selection and Reliability
1.1 Certification Context
Key Takeaways Explain This Topic AI
The current Pearson code is CCAR-F: 60 questions, 120 minutes, $125, and a passing score of 720 on a 100-1000 scale.
The exam validates that a solution architect can design and deploy production Claude applications, not consumer chatbot usage.
Anthropic committed an initial $100 million to the Claude Partner Network, with free membership for qualifying organizations.
Pearson VUE lists four Claude certification exams (CCAO-F, CCDV-F, CCAR-F, and CCAR-P), open only to Claude Partner Network organizations.
Always verify cost, retake, and proctoring rules in the official portal, because partner-program logistics change quickly.
Last updated: October 2026
What CCAR-F Actually Tests
Claude Certified Architect - Foundations now uses the Pearson exam code CCAR-F ; older materials and this route use CCA-F . It launched on March 12, 2026 with the Claude Partner Network (CPN) . The exam validates that a solution architect can design, scope, and deploy production applications built on Claude; it is not a consumer chatbot-usage test.
The questions are scenario-driven. You are given a business situation ("a support team needs to route 10,000 tickets a day with an audit trail") and asked to pick the design that best balances reliability, cost, latency, and safety. The correct answer is the architecture trade-off that fits the stated constraints, which is why memorizing syntax fails and this guide teaches design reasoning.
Exam Logistics at a Glance
The current Anthropic certification page and Pearson program page control logistics; confirm them again before scheduling.
- Item Detail
- Current code CCAR-F (formerly CCA-F)
- Format Multiple choice and multiple response
- Questions 60
- Time limit 120 minutes (~135 minutes seat time)
- Scoring scale 100 to 1000 (scaled)
- Passing score 720
- Registration fee $125 USD
- Validity 12 months
- Delivery Online proctored or Pearson VUE test center
- Prerequisite None formally required; partner-network access applies
- Retakes 14 / 30 / 90-day waits; four attempts per rolling 12 months
A 720 of 1000 scaled score does not mean "72% of questions correct." Anthropic does not publish a raw-score conversion or public pass rate.
The Five Exam Domains and Their Weights
CCA-F is organized into five competency domains. Their relative weight tells you exactly where to spend study time: Agentic Architecture is the single largest domain, so the agent-versus-workflow decision, agentic loops, and multi-agent orchestration deserve the most attention.
- # Domain Weight This guide's coverage
- 1 Agentic Architecture & Orchestration 27% Chapter 2
- 2 Tool Design & MCP Integration 18% Chapter 4
- 3 Claude Code Configuration & Workflows 20% Chapter 3
- 4 Prompt Engineering & Structured Output 20% Chapter 5
- 5 Context Management & Reliability 15% Chapter 6
Note that Domains 2, 3, and 4 together account for 58% of the exam, but no single one dominates. Domain 1 at 27% is the heaviest, and Domain 5 at 15% is the lightest yet still meaningful. A balanced study plan weights practice roughly to these percentages rather than treating every topic equally.
Where CCA-F Sits in the Program
Foundations is one of four Claude certification exams. Pearson VUE's Anthropic program page lists:
- Claude Certified Associate - Foundations (CCAO-F)
- Claude Certified Developer - Foundations (CCDV-F)
- Claude Certified Architect - Foundations (CCAR-F) , this exam
- Claude Certified Architect - Professional (CCAR-P)
Certification is open only to people at Claude Partner Network (CPN) organizations, and registration in the Anthropic Partner Academy requires a partner company email address on a recognized domain. CPN membership itself is free for qualifying organizations and provides training, technical support, and joint market development.
The Architecture Lens to Read Every Question With
The exam rewards a consistent mental checklist. For each scenario, ask:
What does the user actually need? Distinguish the outcome from the implementation.
What does failure cost? A wrong medical summary is catastrophic; a wrong movie suggestion is not. Cost-of-error sets the reliability bar.
Is the path predictable? Predictable, repeatable tasks favor a workflow (orchestrated by code). Open-ended tasks needing dynamic decisions favor an agent .
What data must be fetched, and from where? This drives retrieval and tool design.
Which tool permissions are strictly necessary? Apply least privilege .
How will the team know it still works after a model, prompt, or integration change? This is the evaluation and monitoring requirement.
Common Traps
Choosing the most powerful or most autonomous design "to be safe" — the exam usually penalizes unnecessary autonomy and cost.
Treating CCAR-F like a coding test and memorizing endpoint parameters; the exam tests design choices , not syntax.
Using launch-era logistics — the current exam is $125, valid for 12 months, and delivered through Pearson VUE.
This guide teaches the principles and trade-offs above. It does not reproduce live exam wording, private partner materials, or proprietary questions.
The Claude Ecosystem You Are Being Tested On
CCA-F assumes fluency in the building blocks Anthropic ships, because production architecture means choosing among them correctly. The exam guide's technology list covers this surface area:
The Claude API - the Messages API with system prompts, multi-turn history, and tool use : Claude stops with stop_reason set to tool_use , your code runs the tool, and the result goes back for the next turn. tool_choice decides whether a tool call is optional, required, or forced.
The Claude Agent SDK - Anthropic's framework for building agents: agentic loops, subagents (spawned with the Task tool in the exam guide; current docs call it the Agent tool), hooks such as PostToolUse, and session resumption and forking.
Claude Code - the CLAUDE.md hierarchy, .claude/rules/ , custom slash commands and skills, plan mode, and non-interactive -p runs in CI with --output-format json and --json-schema .
Structured output - tool use with JSON schemas (required, optional, nullable, and enum fields) so downstream code can parse results reliably, plus validation-retry loops for errors a schema cannot catch.
The Message Batches API - 50% cost savings and up to a 24-hour processing window, with custom_id to match requests to results, for latency-tolerant work only.
The Model Context Protocol (MCP) - the open standard for connecting Claude to external data and tools through a uniform server interface, configured in a project .mcp.json or a personal ~/.claude.json .
The guide also lists what will not appear: fine-tuning, API authentication and billing, deploying or hosting MCP servers, streaming implementation, rate limits and pricing, model comparison metrics, embeddings and vector databases, computer use, vision, and prompt caching details beyond knowing the feature exists.
Why "Architecture" Is the Operative Word
A developer asks "how do I call the API?" An architect asks "given this volume, budget, latency SLA, and risk profile, which combination of model, pattern, retrieval, and guardrails best serves the user?" CCA-F is written for the second question. Expect distractor answers that are technically functional but mismatched to the stated constraints - for example, an autonomous agent proposed for a task that a two-step prompt chain would handle more cheaply and predictably. Your job on each item is to select the design whose trade-offs align with the scenario, then be able to defend why the alternatives are worse.