Home › Certifications › CCAR-F
Claude Certified Architect, Foundations (CCAR-F)
An independent community resource - not affiliated with or endorsed by Anthropic. Facts on this page are transcribed from the official exam guide and the Anthropic Partner Academy; the practice questions are written from scratch and none come from a real exam.
Claude Certified Architect, Foundations (exam code CCAR-F) is the first technical exam in Anthropic's Claude certification program and the oldest of the four. Until July 2026 it was the only Claude exam that existed. It is built around scenarios instead of separate questions, and it tests exact details more than any other exam in the program: command options, file paths, parameter values.
The name confuses people. “Foundations” describes the level of detail, not the level of difficulty. The exam sits below a “Professional” exam that many people find easier. All facts below come from the official exam guide, version 1.0 (July 2026), and from the Anthropic Partner Academy certification FAQ.
Exam at a glance
- Exam code
- CCAR-F
- Questions
- 60. Multiple choice and multiple response. Each question says how many answers to pick.
- Format
- 4 scenarios picked at random from a set of 6, about 15 questions each
- Time limit
- 120 minutes. Plan for about 135 minutes in the seat, including check-in and a short survey.
- Passing score
- 720 on a scale of 100 to 1,000
- Valid for
- 12 months
- Retakes
- 14 days after a first fail, 30 after a second, 90 after a third. Up to 4 attempts per rolling 12 months.
- How it is delivered
- Proctored. Online with a webcam, or at a Pearson VUE test center.
- Prerequisites
- None. The guide suggests 6+ months of hands-on building with Claude.
- Language
- English only. Browser translation tools are not allowed during the exam.
- Exam guide
- Version 1.0, effective July 2026
- Practice questions here
- 360, free, every answer option explained
The exam is closed book. No notes, no documentation, no AI assistants. The score report shows the scaled score, the pass or fail result, and the percent correct in each section.
Who it is for
The guide describes a solution architect who designs and builds production applications on Claude, with hands-on experience across the Agent SDK, Claude Code, MCP and the API. It suggests at least six months of real building work.
Daily hands-on use matters more here than in the other three exams, because much of what is tested is detail you normally look up rather than remember.
The scenario format
CCAR-F is the only exam in the program built around scenarios. Six exist. You get four of them, picked at random, with about fifteen questions each. All six are printed in the guide, so you can read them before the exam:
- Customer support agent: Agent SDK, custom MCP tools (get_customer, lookup_order, process_refund, escalate_to_human), a target of solving 80% of cases on first contact, and when to hand a case to a human
- Writing code with Claude Code: custom slash commands, CLAUDE.md, plan mode against running directly
- Multi-agent research system: one lead agent handing work to subagents for search, analysis, summary and reports
- Developer productivity: exploring unfamiliar code with the built-in tools (Read, Write, Bash, Grep, Glob) plus MCP servers
- Claude Code in CI/CD: automatic code review, writing tests, comments on pull requests, and cutting false alarms
- Pulling data out of documents: JSON schema checks, odd cases, and passing results to other systems
Because you get four out of six, about a third of the possible content will not appear on your exam. You cannot know which third in advance, so all six need preparation. One side effect: two people can sit noticeably different exams, so single reports about difficulty do not generalize well.
What the exam covers
| # | Domain | Weight | ≈ questions | In the bank here |
|---|---|---|---|---|
| 1 | Agentic Architecture and Orchestration | 27% | ~16 | 96 |
| 3 | Claude Code Configuration and Workflows | 20% | ~12 | 72 |
| 4 | Prompt Engineering and Structured Output | 20% | ~12 | 72 |
| 2 | Tool Design and MCP Integration | 18% | ~11 | 66 |
| 5 | Context Management and Reliability | 15% | ~9 | 54 |
| Total | 100% | 60 | 360 |
Agent design at 27% is the largest single domain in any of the four exams. It also goes into real
detail. The guide names specific things to know. How to run an agent loop from
stop_reason values instead of reading the reply text for hints that it is done. That
subagents do not see the lead agent's history. How to start subagents with the Task tool,
which needs “Task” listed in allowedTools. How to send several Task calls
in one reply so they run at the same time. And when to use hooks instead of instructions in a
prompt: a hook gives a guarantee, a prompt only gives a good chance.
Claude Code and structured output are 20% each. For Claude Code, know the settings order: user,
then project, then folder CLAUDE.md. Then @import, .claude/rules/ with
path patterns, .claude/commands/, and .claude/skills/ with
context: fork and allowed-tools. Also plan mode against running directly,
and the command line options for CI.
For structured output, know tool_use with JSON schemas. Know the difference between
tool_choice set to “auto”, “any” or a named tool. Know how
optional fields stop the model inventing values, and when to add few-shot examples for unclear
cases.
The technology list and what is ruled out
The appendix of the guide lists the technology that can appear. This is the closest thing to a syllabus, and most people never open it.
- Claude Agent SDK: agent definitions, agent loops, stop_reason handling, hooks (PostToolUse, blocking a tool call), starting subagents with Task, allowedTools
- MCP: servers, tools, resources, the isError flag, writing tool descriptions, splitting tools across agents, .mcp.json, environment variables in config
- Claude Code: CLAUDE.md order, .claude/rules/ with path patterns, slash commands, SKILL.md settings, plan mode, /memory, /compact, --resume, fork_session, the Explore subagent
- Claude Code command line:
-p/--print,--output-format json,--json-schema - Claude API: tool_use with JSON schemas, tool_choice options, stop_reason values, max_tokens, system prompts
- Message Batches API: 50% cheaper, up to 24 hours to finish, custom_id to match requests to answers, no tool calling back and forth
- JSON Schema and Pydantic: required and optional fields, enums, empty values allowed, strict mode, retry loops when checks fail
- Built-in tools: Read, Write, Edit, Bash, Grep, Glob, and when to pick which
- Context and sessions: token budgets, shortening history as you go, the fact that models miss things in the middle of long text, scratchpad files, named sessions, fork_session, keeping contexts separate
The list of ruled-out topics is just as useful, because it tells you what to stop studying. The guide says these will not appear: training or fine-tuning models; API logins, billing or account settings; deep detail of specific programming languages and frameworks; hosting MCP servers (servers, networking, containers); how Claude is built or trained; Constitutional AI, RLHF and safety training; embedding models and vector databases; computer use; image analysis; how to build streaming; rate limits, quotas and price math; OAuth and key rotation; specific AWS, GCP or Azure setup; benchmarks; how prompt caching works under the hood, beyond knowing it exists; and how tokens are counted.
Does it count for your employer?
Yes. Anthropic's certification FAQ lists Architect Foundations as one of the three exams that count toward Claude Partner Network eligibility, together with Developer Foundations and Architect Professional. Claude Certified Associate is the one exam that does not count.
Certification adds to your company's standing in the Claude Partner Network and signals to customers that the team has built a real Claude practice. Companies can see their total number of certifications in the Partner Portal, updated weekly.
One practical rule. You must register with a work email on a domain that is recognized on your company's partner record. A certification earned on an unrecognized domain, including a subsidiary or regional domain, will not be credited to the partner account. Domain changes are made by partner-support@anthropic.com and take 7 to 10 days, so ask before you book.
Preparation checklist
1. Start here
- Create an Anthropic Partner Academy account (needs a work email at a partner company)
- Read the official CCAR-F exam guide in full: 60 questions, 120 min, pass 720/1000, valid 12 months, retakes after 14/30/90 days
- Read all 6 scenarios in the guide - you can see them before the exam
- Read the ruled-out topics list in the appendix so you stop studying them
- Open the prep course list (no single path for this exam, separate courses)
2. Study the 5 exam domains
- Domain 1 - Agentic Architecture and Orchestration (27%, ~16 questions, the biggest)
- Domain 2 - Tool Design and MCP Integration (18%, ~11 questions)
- Domain 3 - Claude Code Configuration and Workflows (20%, ~12 questions)
- Domain 4 - Prompt Engineering and Structured Output (20%, ~12 questions)
- Domain 5 - Context Management and Reliability (15%, ~9 questions)
3. Official courses worth taking
4. Hands-on practice - this exam tests exact details
- Build an agent loop driven by stop_reason, not by reading the reply text
- Set up CLAUDE.md at user, project and folder level, then run /memory to check
- Write a skill with SKILL.md frontmatter: context fork and allowed-tools
- Add a hook that blocks a destructive action, and watch it fire
- Run the CLI headless: -p, --output-format json, --json-schema
- Use tool_choice set to auto, any and a named tool, and see the difference
- Send one job through the Message Batches API and match answers by custom_id
5. Practice tests
360 CCAR-F practice questions, free
Domain quizzes, quick checks and the full 60-question timed simulator, weighted to the official blueprint. Every answer option carries its own explanation of why it is - or is not - correct, so a lucky guess reads the same as a wrong answer.
Open the CCAR-F question bank → Try 5 questions without an account →
The question bank is free. It asks for an account only because the quiz engine has to store answers to score them and show which domains are weak. The 5-question sample needs nothing.
Note on official practice material. The practice exam that ran on the old platform was retired when the program moved to Pearson VUE. Anything now shared as an “official practice test” is a copy of the retired one. The exam guide's sample questions are the only official example of question style.
Registration, retakes and rules
Who can register. Certification is open only to people at Claude Partner Network organizations. Registration needs a partner work email on a recognized company domain. Personal email addresses do not work, and there is no public registration route.
Booking. Register and pay on the Partner Academy first, then use the emailed link to schedule with Pearson VUE, either online or at a test center. A registration stays valid for 5 years, so you can pay now and sit the exam later.
Price and discounts. Discounts follow your company's partner tier. Registered-tier partners pay full price. Select, Preferred and Global Premier partners get 50% off automatically at checkout. Through 31 August 2026, Global Premier partners get 100% off; after that date the standard 50% applies. Discounts also apply to retakes.
Cancelling. You can reschedule or cancel free of charge up to 24 hours before your appointment, with a full refund. Inside 24 hours, or if you do not show up, you lose the fee. Cancelling in Pearson VUE only frees the time slot; it does not issue a refund. To get money back, email certifications-support@anthropic.com.
Retake rules.
- Fail once, wait 14 days
- Fail twice, wait 30 days
- Fail three times, wait 90 days
- Up to 4 attempts per exam per rolling 12 months
- You pay the full fee every time, minus your partner discount
- Limits are per exam. Failing this one does not block a different one.
If you attempted this exam before 30 June 2026 and did not pass, that earlier failed attempt was cleared in the move to Pearson VUE. You can register again with no waiting period.
On exam day. Bring a valid, unexpired, government-issued photo ID. The name must match your registration exactly. To fix a name, email certifications-support@anthropic.com before you book. Your desk must be clear: no phone, watch, headphones, notes or second screen. Run Pearson's system test on the computer you will use. If a managed work device blocks the exam, take it at a test center instead. Ask for accommodations at least 10 days ahead. Full rules are in the certification policies.
Results and badge. You see your score on screen at the end. If you pass, Credly emails you an invitation to claim the badge, usually within minutes for online exams. Adding a personal email to your Credly profile keeps your badges if you change jobs.
Renewal. The credential lasts 12 months. Renewing on time means a short free assessment on the Partner Academy, with no proctor and no fee. If you let it lapse, you sit the full exam again at full price.
Appeals. You can appeal within 14 days of being told about a decision, or within 14 days of your exam date for a concern about a result. Appeals go to Pearson VUE. Reporting a problem with a question never counts against you, but one disputed question does not by itself change a pass or fail.
References
- Claude Certified Architect, Foundations Exam Guide, version 1.0, July 2026 (exam code CCAR-F)
- Anthropic Partner Academy, certification FAQ
- Anthropic Partner Academy, certification policies
- CCAR-F certification page and exam guide
- Pearson VUE, Anthropic exam support
- Anthropic product documentation: Agent SDK, Claude Code, MCP, Messages and Message Batches APIs
Help keep this current
Exam guides get reissued, scenarios get swapped, and the people who know what a sitting is actually like are the ones who just had one. This page is meant to keep changing.
Already passed? Send what was worth knowing: a weighting that felt off, a topic the guide underplays, a detail that mattered on the day. Not exam questions - sharing those breaks the agreement you signed at registration, and it would make this page worthless as preparation anyway. Everything else helps whoever sits it next.
Still preparing? What is on this page comes from the official guide and from people who have actually sat the exam, not from a scrape. If something is unclear, or you hit something the guide does not answer, ask - the answer comes back onto the page, so the next reader gets it too.
Either way: admin@termidy.com.
This page is written from the official exam guide (version 1.0, effective July 2026) and the Anthropic Partner Academy FAQ, with input from people who have taken the exam.
Stay current - and take the video course free
We follow AI for our own work and pass on only what changes how you build: certification news, policy shifts, new models. Signing up also makes the Architect Foundations video course free.