Claude Certified Architect: What the Exam Covers

Claude Certified Architect – Foundations, exam code CCAR-F, is Anthropic’s certification for solution architects who design and build production applications with Claude. It tests judgment across four technologies, Claude Code, the Claude Agent SDK, the Claude API and the Model Context Protocol, through scenario-based questions: 60 items in 120 minutes, drawn from 4 of 6 real-world scenarios, with a scaled passing score of 720 out of 1,000. Pearson VUE delivers it online or at a test center, and the credential lasts 12 months. The catch for most readers: as of September 30, 2026, only people at companies in the Claude Partner Network can register. This guide sets out what the exam covers as Anthropic’s own exam guide describes it, and how to prepare.

Who can take it

Anthropic first announced the credential when it launched the Claude Partner Network↗ (opens in a new tab) on March 12, 2026, as “the first Claude technical certification.”

Since then it has become one of four role-based credentials, and Anthropic’s certification announcement↗ (opens in a new tab) of July 23, 2026 states that all of them are available to members of the Claude Partner Network only. In practice that means:

• Your employer must be a Claude Partner Network member. Membership is free, and any organization bringing Claude to market can apply.
• You register with your company email on a domain your partner record recognizes. Personal addresses are refused.
• You must be at least 18, and you show a government-issued photo ID whose name matches your registration exactly.
• There is an exam fee for each attempt, with discounts by partner tier. This guide does not quote it; check the exam page.

The four credentials are Associate – Foundations for everyday use of Claude on projects, Developer – Foundations for engineers building with the API, Architect – Foundations, and Architect – Professional for enterprise-scale integration, governance and evaluation. Foundations is not a prerequisite for Professional, and it does not upgrade into it.

Who it is written for

The CCAR-F exam guide↗ (opens in a new tab) describes the ideal candidate as a solution architect with about six months or more of hands-on work with the Claude API, the Agent SDK, Claude Code and MCP. It expects you to have built agentic applications with subagents and hooks, configured Claude Code for a team with CLAUDE.md, skills, MCP servers and plan mode, designed MCP tools, engineered prompts for reliable structured output, managed long contexts, wired Claude into CI/CD, and made escalation and human-in-the-loop decisions.

The format at a glance

• 60 multiple-choice and multiple-response items; each item says how many answers to pick.
• 4 scenarios, picked at random from a bank of 6, each framing a set of questions.
• 120 minutes. Anthropic’s FAQ suggests planning about 135 minutes of seat time with check-in and a short survey.
• Closed book: no notes, no documentation, no translation tools and no AI assistants.
• Scaled score from 100 to 1,000; 720 passes. The score report adds your percentage correct in each domain.
• Valid for 12 months. Renewing on time means reviewing what changed and passing a free, unproctored assessment; a lapsed credential means sitting the full exam again.

What the exam covers: five domains

1. Agentic Architecture & Orchestration, 27%: agentic loops driven by stop_reason (tool_use versus end_turn), coordinator and subagent patterns, passing context to subagents explicitly, hooks that block or rewrite tool calls, task decomposition, and resuming or forking sessions.
2. Tool Design & MCP Integration, 18%: tool descriptions that keep similar tools apart, structured errors with the MCP isError flag, spreading tools across agents, MCP server configuration in Claude Code, and choosing among the built-in tools.
3. Claude Code Configuration & Workflows, 20%: the CLAUDE.md hierarchy, custom commands and skills, path-scoped rules in .claude/rules/, plan mode versus direct execution, iterative refinement, and running Claude Code in CI with -p and JSON output.
4. Prompt Engineering & Structured Output, 20%: explicit criteria, few-shot examples, structured output through tool use and JSON schemas, validation and retry loops, batch processing, and multi-pass reviews.
5. Context Management & Reliability, 15%: keeping critical facts through long conversations, trimming tool output, escalation, error propagation, confidence calibration and keeping track of where each claim came from.

The six scenarios

• A customer support resolution agent built on the Agent SDK, with MCP tools for customers, orders, refunds and escalation to a human.
• Code generation with Claude Code, using custom commands, CLAUDE.md and plan mode.
• A multi-agent research system in which a coordinator delegates search, analysis, synthesis and reporting to subagents.
• Developer productivity tools built on the Agent SDK with the built-in tools and MCP servers.
• Claude Code in continuous integration: automated reviews, test generation and pull request feedback with few false positives.
• Structured data extraction from unstructured documents, validated against JSON schemas.

What is not on it

The guide lists topics that will not appear. Among them: fine-tuning, API authentication, billing and account management, API key rotation and OAuth details, rate limits and pricing, streaming, computer use, vision, cloud-provider setup, embeddings and vector databases, tokenization, and model training methods. Those are worth knowing for real work, and some have their own guides, such as keeping Claude API keys safe, but they will not earn you points.

Where the guide and the docs differ

Two small gaps are worth knowing before exam day. First, the guide calls the subagent-spawning tool the Task tool and says allowedTools must include "Task". Claude Code’s subagents documentation↗ (opens in a new tab) says the Task tool was renamed Agent in version 2.1.63, with Task(...) still accepted as an alias. Expect exam items in the guide’s vocabulary. Second, the guide says you may cancel or reschedule up to 24 hours before your appointment, while the certification FAQ and Pearson VUE’s Anthropic page both say 48 hours. Plan on 48.

How to prepare

The guide’s own advice is to build things, and it names what: an agent with a full tool-calling loop, error handling, sessions and subagents; a real project’s Claude Code setup with a CLAUDE.md hierarchy, path rules, skills and at least one MCP server; MCP tools whose descriptions hold up under ambiguous requests; an extraction pipeline with JSON schemas, validation retries and the Message Batches API; few-shot prompts and multi-pass reviews; and escalation rules for when an agent should hand over to a person. It also includes four hands-on exercises and sample questions with explanations.

• Official prep: each certification has prep courses in the Anthropic Partner Academy, free to partners.
• Public courses: Building with the Claude API, Introduction to Model Context Protocol, Claude Code in Action, and the agent skills and subagents courses on Claude Academy cover much of the ground; see Anthropic Academy for the list.
• Reading: Claude Code best practices, the Claude Agent SDK, MCP tool descriptions and MCP error handling map onto domains 1 to 3.
• Practice on your own work. The scenarios are built from customer use cases, so a real agent you have debugged teaches more than any summary.

Registering, exam day and retakes

You register and pay on the exam’s page in the Anthropic Partner Academy, then schedule through Pearson VUE’s Anthropic page↗ (opens in a new tab), online with OnVUE or at a test center. If you do not pass, the waits grow with each attempt: 14 days after the first, 30 after the second and 90 after the third, with at most four attempts per exam in any rolling 12 months, each at the full fee. A pass sends a digital badge through Credly that you can add to LinkedIn.

Keep the study plan where you can see it

The guide’s exercises and task statements make a ready-made backlog. Paste them into a board with Add many and each becomes a task you move from To Do through Next Up and In Progress to Completed, with the note holding what the objective asks and the plan holding the small build that will prove you can do it. In fenbs, an AI assistant connected over MCP can read that board and quiz you on the task in progress, and every change is recorded with who made it. fenbs has no due dates, so keep the exam date on your calendar, not on the board.