# CCAR-F 2026 Guide: Claude Certified Architect, Free Practice

> Claude Certified Architect (CCAR-F) explained: 60 questions, five weighted domains, who it suits, what it costs, and how independent engineers become eligible.

Source: https://youraidept.com/network/ccar-f

Exam guide

# The Claude Certified Architect – Foundations (CCAR-F) exam.

CCAR-F is Anthropic's foundations certification for engineers and solution architects who design Claude-powered systems. It is the flagship exam of the program, often shortened to CCA-F, and the one most independent AI engineers sit first.

## What is CCAR-F?

CCAR-F (Claude Certified Architect – Foundations) is Anthropic's certification for engineers and solution architects who design Claude-powered systems. The exam is 60 questions in 120 minutes, proctored and closed-book, delivered by Pearson VUE, with a passing score of 720 on a 100–1,000 scale and a $125 fee per attempt.

## CCAR-F at a glance

### Claude Certified Architect – Foundations

- Questions: 60 items in 120 minutes
- Passing score: 720 on a 100–1,000 scale
- Exam fee: $125 per attempt, per the current exam guide
- Format: Proctored and closed-book, no AI assistance
- Delivery: Pearson VUE, online proctored or at a test center
- Retakes: Up to 4 attempts per rolling 12 months (14, 30, then 90-day waits)
- Validity: 12 months, with a free on-time renewal assessment

## What the exam covers

Five domains, per the published exam guide.

| Domain | Weight |
| --- | --- |
| Agentic architecture | 27% |
| Claude Code workflows | 20% |
| Prompt engineering | 20% |
| Tool design and MCP | 18% |
| Context management | 15% |

## Try three CCAR-F practice questions

One from each of the heaviest domains, with the answer and the reasoning. Every question in our bank is original, written from the published blueprint, never taken from the exam.

Agentic architecture

A support desk needs incoming tickets sorted into five fixed categories and routed to the matching queue. Which design fits best?

- An orchestrator that spawns five subagents and lets each one claim the tickets it judges to be in scope
- An evaluator-optimizer loop that proposes a category and revises it until a second judge call agrees
- A routing workflow with one classification call, then a fixed handoff to the queue for that category
- A single autonomous agent with the queue tools attached, deciding per ticket where it should go

Answer: A routing workflow with one classification call, then a fixed handoff to the queue for that category. The control flow is known in advance and the output space is fixed, so a routing workflow (classify, then hand off) is cheaper, faster, and easier to test. The evaluator-optimizer loop adds iterations to a decision that has nothing to refine, and an orchestrator with five subagents spends five calls to make one choice.

Claude Code workflows

What belongs in a project's CLAUDE.md file?

- The full architecture documentation and API references, so Claude has complete knowledge in every session
- Environment variables and service credentials the project needs, so Claude can run commands without asking
- Build and test commands, project conventions, and gotchas Claude needs on every task, kept short
- A running log of past sessions and their outcomes, so Claude can pick up where the last one ended

Answer: Build and test commands, project conventions, and gotchas Claude needs on every task, kept short. CLAUDE.md is standing project memory: short, high-value instructions loaded into every session. Long documents cost context on every task and are better read on demand; secrets do not belong in a checked-in file; session history is what memory files or notes are for.

Prompt engineering

You need a response in a strict JSON shape that downstream code will parse. Which approach is most reliable?

- Ask for JSON in the user message and add a note that any text outside the object will break the parser
- Define the shape as a tool input schema, or use structured outputs, so the model must conform
- Lower the temperature to zero so the formatting is deterministic across runs
- Prefill the assistant turn with an opening brace and stop at the closing brace

Answer: Define the shape as a tool input schema, or use structured outputs, so the model must conform. Schema-enforced output is more reliable than instructions alone: tool use with an input schema, or structured outputs, constrains the shape. Prefilling the brace suppresses preamble but does not guarantee the keys, types, or nesting downstream code expects.

More than 90 people have started the free CCAR-F practice exam since September, and more than 60 finished a full timed mock. More than 220 engineers have joined the YAID Network. Counts as of October 15, 2026. Keep going with [all 200 free CCAR-F practice questions](https://youraidept.com/network/ccar-f-practice-questions) or the [timed CCAR-F mock](https://youraidept.com/network/ccar-f-practice-exam).

## Who it’s for

Engineers and solution architects who design systems on Claude: agent architectures, tool integrations, and the context strategies that hold them together. If you sell or deliver Claude implementation work, this is the credential clients will recognize first.

## How CCAR-F compares

Against the rest of the program: CCDV-F tests hands-on application code where CCAR-F tests system design, CCAO-F is for non-developers, and CCAR-P is the professional tier that builds on this exam. If you sell or lead Claude implementation work, CCAR-F is the one to sit first; it is also the credential the market shortens to CCA-F.

## How to register

Registration runs through the Anthropic Partner Academy, and access is open to organizations in the Claude Partner Network. There is no individual sign-up. If your employer is not a partner, the practical route is joining a partner firm: YAID members register under our partner standing, prepare with the official Partner Academy materials, and sit the exam through Pearson VUE. The credential is issued to you personally and stays with you. The full path is in [the Claude certification guide](https://youraidept.com/network/claude-certification).

## Preparing for CCAR-F

- Study against the blueprint, not vibes. Agentic architecture alone is 27% of the exam, so weight your prep the way the guide weights the domains.
- The exam is closed-book with no AI assistance. Anything you normally look up needs to be working knowledge before you sit it.
- Candidates who pass tend to recommend a stretch of theory first, then hands-on building. The Partner Academy prep course follows the same blueprint.

Test yourself with the [free timed CCAR-F practice exam](https://youraidept.com/network/ccar-f-practice-exam) or the [CCAR-F practice questions by domain](https://youraidept.com/network/ccar-f-practice-questions), both written from the blueprint and weighted the way the exam is. For a week-by-week schedule, read [a three-week CCAR-F plan built on the blueprint](https://youraidept.com/blog/how-to-prepare-for-the-ccar-f-exam).

## What each domain covers

What to know before you sit, organized by the published blueprint. Nothing here describes an exam question.

- ### Agentic architecture The agent loop: the model reasons, calls a tool, receives the result, and decides whether to continue or stop. Know what ends the loop (a final answer, a stop condition, a max-turns limit, a human checkpoint).
- Workflows versus agents. Anthropic's own framing: prompt chaining, routing, parallelization, orchestrator-workers, and evaluator-optimizer are workflows with fixed control flow; an agent decides its own path. Be able to say which pattern fits a given scenario and why a simpler one is often the right answer.
- Orchestrator and subagent designs: what the coordinator keeps, what it delegates, and why subagents get a clean context and a complete brief rather than a pointer to the parent's memory.
- Failure handling: retries with backoff, idempotent tool calls, timeouts, and what the system does when a tool returns an error instead of a result.
- Guardrails and human-in-the-loop: where approval gates belong (spend, deletion, external sends), and how hooks or permission checks enforce them outside the model's judgment.
- Cost and latency trade-offs: model choice per step, parallel tool calls, caching, and when a deterministic function beats a model call.

- ### Claude Code workflows CLAUDE.md as project memory: what belongs in it (commands, conventions, gotchas) and what does not (long documents the model can read on demand).
- The explore, plan, code, commit rhythm from Anthropic's best-practices guide, and plan mode as a way to separate thinking from editing.
- Slash commands, skills, and subagents: reusable prompts versus delegated work with its own context.
- Hooks: deterministic shell steps that run before or after tool calls, used for formatting, linting, blocking dangerous commands, and notifications.
- Permission modes and allow lists, and why the default asks before writes and shell commands.
- MCP server configuration at project and user scope, and headless use in CI and scripts with the print flag.
- Context hygiene: clearing and compacting, and why long sessions degrade without it.

- ### Prompt engineering System prompts for role and standing instructions; the user turn for the task. Clear, specific instructions work better than clever ones.
- Examples: a small number of high-quality examples steer format and tone more reliably than description alone.
- Structure with XML tags to separate instructions, documents, and examples, and to make long prompts unambiguous.
- Reasoning: chain-of-thought instructions and extended thinking, and when each is worth the tokens.
- Controlling output: response prefilling, requested formats, JSON via tool definitions or structured outputs.
- Long-context habits: put documents at the top, ask for quotes before analysis, and keep the question at the end.

- ### Tool design and MCP A tool definition is a name, a description, and a JSON schema for inputs. The description is what the model reads to decide when to call it; write it for the model, with edge cases and examples.
- A few well scoped tools work better than many overlapping ones. Namespace related tools, return meaningful context rather than raw dumps, and keep responses within a sane token budget.
- The tool_choice parameter (auto, any, a specific tool, none), parallel tool use, and how a tool result with an error flag differs from a successful result.
- MCP primitives: tools, resources, and prompts. Servers expose them; clients (Claude Code, the Agent SDK, Claude.ai) connect to them.
- MCP transports: stdio for local processes, streamable HTTP for remote servers, and where authentication sits for remote ones.
- Idempotency and safety: which tools need confirmation, which are read-only, and how the schema and description communicate that.

- ### Context management The context window is a budget: system prompt, conversation, tool definitions, tool results, and the response all share it.
- Context rot: quality falls as the window fills with stale material. Compaction, summarization, and clearing exist to fight it.
- Load context when it is needed rather than up front. Give the agent lightweight identifiers (file paths, record ids, URLs) and let it fetch what it needs.
- Structured note-taking and memory files as persistence across sessions; subagents as a way to isolate context for a subtask and return only the conclusion.
- Prompt caching: what makes a prefix cacheable, why stable system prompts and tool definitions go first, and the cost effect.
- Retrieval versus stuffing: when to search and load a slice versus pasting a whole corpus, and how chunking and citations interact.

## Common mistakes

- Reaching for an agent when a workflow with fixed control flow would be cheaper, faster, and easier to test. The exam rewards the simplest design that meets the requirement.
- Writing tool descriptions as if a person will read them. The model reads them to decide when to call the tool, so the description has to carry the edge cases and an example or two.
- Ignoring the closed-book format. People who build with Claude daily still lose marks on parameter names and policy details they normally look up.
- Spreading prep evenly across the five domains when agentic architecture carries 27% on its own.

## A study path

- 01 Week 1: read the exam guide end to end and the four Anthropic engineering posts linked below. Write a one-page summary of each domain in your own words.
- 02 Week 2: build a small agent with the Agent SDK or Claude Code that uses at least two tools and one MCP server. Add a hook and a permission rule.
- 03 Week 3: work through the Partner Academy prep course, then a community practice exam. Review every miss against the docs.
- 04 Week 4: closed-book review. Rewrite your domain summaries from memory and compare.

### Official reading

- [Building effective agents](https://www.anthropic.com/engineering/building-effective-agents)
- [Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)
- [Writing tools for agents](https://www.anthropic.com/engineering/writing-tools-for-agents)
- [Claude Code best practices](https://www.anthropic.com/engineering/claude-code-best-practices)
- [Claude Developer Platform docs](https://docs.claude.com/)
- [Claude Code docs](https://code.claude.com/docs)
- [Model Context Protocol](https://modelcontextprotocol.io/)

## CCAR-F FAQ

### What is the CCAR-F exam? +

CCAR-F is Claude Certified Architect – Foundations, Anthropic's certification for engineers and architects who design Claude-powered systems. It is 60 questions in 120 minutes, proctored and closed-book, with a passing score of 720 on a 100–1,000 scale.

### How much does the CCAR-F exam cost? +

$125 per attempt, per the current exam guide. The fee is paid to the certification program when you register. Program pricing changes, so check the current guide before scheduling.

### Is CCAR-F the same as CCA-F? +

Yes. CCA-F is a common shorthand; the official Pearson VUE listing uses CCAR-F for Claude Certified Architect – Foundations.

### How do I register for the CCAR-F exam? +

Registration runs through the Anthropic Partner Academy, which is open to people affiliated with a Claude Partner Network organization. Independent engineers typically get access by joining a partner firm such as YAID, then schedule through Pearson VUE.

### How hard is CCAR-F? +

It assumes real working knowledge across five domains: agentic architecture, Claude Code, prompt engineering, tool design and MCP, and context management. Engineers who build with Claude regularly report the material is fair; the closed-book format is what catches people who rely on looking things up.

## The other Claude certifications

CCDV-F

### [Claude Certified Developer – Foundations](https://youraidept.com/network/ccdv-f)

CCAO-F

### [Claude Certified Associate – Foundations](https://youraidept.com/network/ccao-f)

CCAR-P

### [Claude Certified Architect – Professional](https://youraidept.com/network/ccar-p)

Not sure the credential is worth the prep time? [The honest case for and against Claude certification](https://youraidept.com/network/claude-certification-worth-it).

## Sources

- [Pearson VUE: Claude Certification Program by Anthropic (exams, eligibility, retake policy)](https://www.pearsonvue.com/us/en/anthropic.html)
- [Anthropic: Services track and Partner Hub announcement](https://www.anthropic.com/news/services-track-partner-hub)

Facts on this page were last checked against the program pages on October 15, 2026. Exam details are Anthropic’s and change over time; fees marked as reported are not yet on an official public page. Verify against the official pages before you rely on any number here.

## Who wrote this

### [Matei Olaru](https://youraidept.com/about/matei-olaru)

CEO, Your AI Department

Former public-company CEO, corporate attorney, and World Bank consultant. Runs the YAID Network and the partner relationship with Anthropic.

### [Michael Greenberg](https://youraidept.com/about/michael-greenberg)

AI Product Lead, Your AI Department

AI at Google, Meta, and Samsung. Six patents. Leads member support and the certification prep guidance.

Your AI Department is a registered Claude Partner Network company. Our members sit these exams under our partner standing. Facts on this page were last checked on October 15, 2026.

## Become eligible to sit CCAR-F.

[YAID Network](https://youraidept.com/network) membership gets you exam registration under our Claude Partner Network standing, plus the firm infrastructure around your client work. $149 per quarter.

[Apply to join YAID ↗](https://youraidept.com/network/apply)[See membership →](https://youraidept.com/network/membership)