The Claude Certified Architect – Foundations (CCA-F) exam measures whether you can make sound architectural decisions when building with Claude — designing agents, wiring up tools, configuring Claude Code, engineering prompts for structured output, and keeping long-running work reliable. It is a reasoning exam, not a trivia quiz: most questions put you inside a realistic scenario and ask which design choice is most correct.
This guide is the map. It explains how the exam is structured, what each of the five domains covers, and how to build a study plan that mirrors the real weighting. From here, each domain has its own deep-dive guide that walks every objective with concrete examples.
This is an independent, unofficial resource. It is not affiliated with, endorsed by, or authorized by Anthropic — it exists to help you reason about the material the public blueprint covers.
What CCA-F is — and who it's for
CCA-F is the entry-level credential in Anthropic's Claude Certified Architect track. It validates that you understand the building blocks of production Claude systems: the agentic loop, tool and MCP design, Claude Code configuration, prompt engineering for structured output, and context/reliability management. The Professional (CCA-P) exam is a separate, more advanced certification — this platform covers Foundations only.
It's aimed at engineers, solutions architects, and technical builders who work with Claude directly — through the Anthropic API, through Claude Code, or by integrating MCP servers. You don't need to have shipped a huge system, but you do need hands-on familiarity with how these pieces fit together. Pure conceptual reading rarely gets people over the line; the scenario questions reward people who have actually built and debugged agents.
How the exam is weighted
The blueprint spans five domains and 29 task objectives, with scenario-based questions drawn from real-world contexts (customer-support agents, multi-agent research, Claude Code in CI/CD, structured data extraction, and more). The domains are not weighted equally — study time should follow the weighting.
| Domain | Weight | Objectives | What it's really about |
|---|---|---|---|
| D1 — Agentic Architecture & Orchestration | 27% | 6 | Designing agents that always resolve or escalate; handoffs; the agentic loop |
| D2 — Tool Design & MCP Integration | 18% | 5 | Tool descriptions, structured errors, MCP scopes, tool_choice |
| D3 — Claude Code Configuration & Workflows | 20% | 8 | CLAUDE.md, slash commands, hooks, permissions, plan mode |
| D4 — Prompt Engineering & Structured Output | 20% | 6 | JSON-schema tool use, extraction schemas, batches vs. sync |
| D5 — Context Management & Reliability | 15% | 4 | Stateless API, subagent isolation, escalation, review routing |
The scoring model — and one honest caveat
The real exam reports a scaled score from 100 to 1000, with 720 to pass, plus a per-domain breakdown showing where you were strong or weak. The scaled score is produced by Anthropic's own item-response weighting, which is not public.
The five domains at a glance
Each domain below has a dedicated deep-dive guide on this site that covers every objective with examples and common traps. Use this section as the index.
- D1 · Agentic Architecture & Orchestration (27%) — Orchestration-layer safeguards, structured handoff packages, session resumption, dynamic task decomposition, PreToolUse/PostToolUse hooks, and how
stop_reasondrives the agentic loop. - D2 · Tool Design & MCP Integration (18%) — Structured tool errors, high-quality tool descriptions, MCP server scopes and auth, and using
tool_choiceto guarantee tool invocation. - D3 · Claude Code Configuration & Workflows (20%) — Custom slash commands, iterative refinement, plan mode, and choosing between CLAUDE.md,
.claude/rules/, Skills, hooks, and permissions. - D4 · Prompt Engineering & Structured Output (20%) — The reliability ladder for structured output, extraction schemas that don't fabricate, few-shot accuracy patterns, and sync vs. Message Batches.
- D5 · Context Management & Reliability (15%) — Subagent isolation, scratchpads, the stateless Messages API, escalation criteria, and confidence-based human review routing.
A four-week study plan
This plan mirrors the exam weighting. Adjust to your baseline — if you already build agents daily, you can compress it.
Exam-day tactics
- Read for 'most correct', not 'correct'. Several options are often defensible; the exam wants the best architectural choice for the stated constraints. Anchor on the scenario's specific requirements.
- Watch for the model-discretion trap. If a question involves a rule that must hold, the answer is almost never 'add it to CLAUDE.md and hope' — it's a hook, permission, or schema.
- Eliminate fabrication. Options that would make the model invent data, leave a session hanging, or ignore an explicit escalation request are usually wrong on principle.
- Don't over-engineer. The simplest design that satisfies the constraints beats a clever one that adds risk. Prefer the answer with fewer moving parts when both are correct.
Key takeaways
- 01CCA-F is Foundations-only; Professional (CCA-P) is a separate exam.
- 02Five domains, 29 objectives, scenario-based; passing is 720/1000.
- 03D1 (27%) is the heaviest domain — invest study time proportionally.
- 04The recurring theme: model discretion vs. deterministic enforcement.
- 05No honest tool predicts a scaled score — study by per-objective accuracy.
- 06Prefer the simplest design that satisfies the stated constraints.
Common mistakes
Studying all five domains equally.
Weight your time to the blueprint — D1 is 27%, D5 is 15%. Spend accordingly.
Chasing a predicted scaled score from practice tools.
No public formula exists. Track raw score and per-objective accuracy instead.
Reading docs without building anything.
The scenario questions reward hands-on experience. Build a small agent, an MCP server, and a hook before exam day.
Reviewing mocks only by total score.
Review by objective. One weak objective can be masked by a strong average.
Frequently asked
How many questions is the CCA-F exam and how long is it?
Anthropic sets the exact question count and time limit, and those details can change — check the official exam page for the current format. What's stable is the blueprint: five weighted domains and 29 objectives, scenario-based, passing at 720/1000.
Do I need to know how to code to pass CCA-F?
You need to read and reason about code and configuration — JSON schemas, tool definitions, hook configs, CLAUDE.md — and understand the architectural trade-offs. You won't be writing large programs, but hands-on familiarity with the Anthropic API, Claude Code, and MCP is close to essential.
Is CCA-F worth it before CCA-P?
Foundations is the entry credential and establishes the shared vocabulary and mental models the Professional exam builds on. Most people take Foundations first.
Does this site cover the Professional (CCA-P) exam?
No. This platform focuses only on Foundations (CCA-F). Professional is a separate, more advanced certification we don't currently provide content for.
Independent, unofficial study material. Not affiliated with, endorsed by, or authorized by Anthropic. Every example is original and written to teach the public exam objectives — no real exam questions are reproduced. Technical details reflect Claude, the Anthropic API, Claude Code, and MCP as of July 24, 2026; always confirm specifics against current official documentation.
Test yourself
Turn what you just read into answers you can check.