CCAR-F certification prep

Build Claude architects who can prove it.

Scenario-based preparation for Anthropic's Claude Certified Architect – Foundations (CCAR-F) exam. For individual candidates, and for firms building a certified Claude practice.

Free to try — Domain 1 is open with no payment required.

Independent preparation based on Anthropic's published exam guide. Not affiliated with or endorsed by Anthropic.

+6 more

Scenario 2 of 4 · Question 17 of 60

Multi-agent research system

A subagent returns a tool error mid-loop. The lead agent still owes the caller a cited report. What should it do with the failure?

  • A. Retry the call unchanged and raise the timeout
  • B. Return the tool error to the model as structured output
  • C. Escalate to a human before any retry
  • D. Swallow the error and continue the loop

Correct. A structured error keeps the loop recoverable and the citation chain intact.

5domains30task statements

Blueprint coverage

weights per Anthropic's published exam guide

  1. D1

    27%

    Agentic Architecture & Orchestration

  2. D2

    18%

    Tool Design & MCP Integration

  3. D3

    20%

    Claude Code Configuration & Workflows

  4. D4

    20%

    Prompt Engineering & Structured Output

  5. D5

    15%

    Context Management & Reliability

30

task statements, every one mapped to its own scenarios

Why this is different

Multiple choice, like the exam. The options are the hard part.

Same format you will be tested in. The difference is that you cannot eliminate three obvious throwaways and move on.

Multi-agent research system

A lead agent hands a research task to a subagent. What should travel with it?

  • A. Put the whole prior conversation in the subagent's prompt so nothing is lost
  • B. Pass a structured summary with source URLs and document names alongside the findings
  • C. Pass only the findings, and let the lead agent re-derive attribution afterwards
  • D. Have each subagent write to shared storage and pass a pointer

Why the others are tempting

None of these are filler. Each one is something a competent engineer might ship.

A

Plausible: it does preserve everything. Wrong: it spends the subagent's context on material it will never use, and the handoff is where budgets are tightest.

C

Plausible: it is the leanest handoff. Wrong: attribution cannot be reconstructed once the source metadata is gone, and the report has to cite.

D

Plausible: it is how you would solve this outside an agent loop. Wrong: it adds a failure mode and an I/O dependency the loop cannot recover from mid-turn.

CCAR Prep.

10 people.
One preparation standard.

Certification is individual. Preparation is the part you can standardise.

  1. 1

    Scope

    Five domains, thirty task statements

  2. 2

    Scenarios

    Production calls in the exam's format

  3. 3

    Evaluation

    The reasoning for every option

  4. 4

    Practice exam

    Sixty items, four scenarios

  5. Ready

    Every candidate to the same standard

Why now

Claude expertise is becoming something firms have to demonstrate.

Anthropic launched the Claude Partner Network in March 2026. Certified practitioners are one of the signals it uses when evaluating the depth of a firm's Claude practice.

Certified bench scale

Registered starts at 10 committed

10

Select

active certified individuals

100

Preferred

active certified individuals

1,000

Global Premier

active certified individuals

Certified-practitioner counts are one component of Anthropic's tier criteria. Partner tiers include additional production deployment and customer-story requirements, and certification counts alone do not earn a tier. Anthropic counts an active certified practitioner as someone holding a current certification who has used Claude in the previous 90 days. Certifications belong to individuals, not firms. Figures from Anthropic's March 12, 2026 Claude Partner Network announcement and its June 3, 2026 Services Track announcement.

Getting 10 certified is not 10 individual problems.

Getting one architect certified is an individual goal. Getting 10, 25, or 100 certified is an operating problem.

The real cost

The exam fee isn't the expensive part.

Without a standard

10

candidates

a starting bench

×

?

hours each

self-directed study

×

?

retry cycles

fee, prep, and a later date

= total preparation cost

Ten people solving the same problem privately, and every avoidable retake adding another fee, another cycle, and a later date for staffing certified people onto client work.

With one standard

Find the gaps before exam day, not after.

  • 01One preparation path, so readiness means the same thing across the team.
  • 02No engineer designing a curriculum out of documentation before they can start.
  • 03Weak reasoning surfaces in practice, which is the cheapest place to find it.

Illustrative. We cannot promise nobody fails, and the fee is the smaller half: the rest is architect time.

Who it's for

Firms building a Claude practice.

AI consultancies

Preparing consultants repeatably.

Development agencies

Architects fluent in production Claude.

Systems integrators

Growing the certified bench.

Internal AI teams

One architecture standard.

Pricing

Which side of this page are you on?

Individual

$5

One time. Lifetime access.

For an architect preparing for CCAR-F themselves.

Start preparing

Free to try — Domain 1 is open before you pay.

Team

Preparing a bench?

Any team size

One consistent preparation path, so readiness means the same thing across everyone you put through it. $100 for a team of ten is a small price for the edge of walking into the exam already prepared. Add teammates one at a time, whenever you're ready.

Start a team — $10

Only $10 per seat.

Already CCAR-F certified?

Continue your path with CCAR-P.

Claude Certified Architect – Professional is the next-level credential for architects who design and own production Claude systems. Domain 1 is free to try.

Certification is individual.
Readiness doesn't have to be.