Claude Certified Architect Exam Domains and Their Weightings

The Claude Certified Architect — Foundations (CCA-F) exam is built on five weighted domains: Agentic Architecture & Orchestration at 27%, Claude Code at 20%, Prompt Engineering at 20%, Tools & MCP at 18%, and Context & Reliability at 15%. Those weights decide how 60 scenario questions are distributed, so they should decide how you spend your study hours.

Plinth Prep is an independent study resource and is not affiliated with or endorsed by Anthropic. The domain names, weightings and pass threshold below come from Anthropic's official exam guide, which is the canonical statement of scope and worth reading in full before you plan anything. The specific figures in this post — 27/20/20/18/15% domain weights, 60 questions, scoring scale of 100–1000, and passing threshold of 720 — were verified against Version 1.0 of that guide, effective July 2026; if any figure differs from the version live at your time of reading, the guide takes precedence.

What are the five Claude Certified Architect exam domains?

Per the official exam guide, the exam is 60 scenario-based multiple-choice questions in 120 minutes, scored on a scale of 100–1000 with a passing score of 720. The five domains and their published weights map onto that 60-question paper as follows:

• Agentic Architecture & Orchestration — 27% (about 16 of 60 questions). Designing multi-step agents, deciding when a model should act rather than answer, orchestration patterns, error propagation, and the guardrails that belong around an agent taking real actions.
• Claude Code — 20% (about 12 questions). Configuration and workflows: how Claude Code is set up, how it is driven day to day, and how it fits into a working development process.
• Prompt Engineering & Structured Output — 20% (about 12 questions). System prompts, few-shot patterns, prompt chaining, and getting reliably structured output instead of prose you then have to parse.
• Tool Design & MCP Integration — 18% (about 11 questions). Function calling at the API level, how to write a tool a model can actually use well, and when the Model Context Protocol is the right integration versus a custom tool definition.
• Context Management & Reliability — 15% (about 9 questions). Context windows, what to keep and what to drop, prompt caching mechanics, and how a system behaves when things degrade.

Add those percentages up and they come to 100. There is no separate retrieval-augmented generation domain and no separate responsible-AI domain on this blueprint. Retrieval and safety appear inside the five domains — retrieval as a context-management problem, safety as a constraint on agent design — which is a meaningfully different thing to study for.

Which domain carries the most weight?

Agentic Architecture & Orchestration, at 27%, is the largest single block on the paper — more than a quarter of your score. Its lead is not marginal: it is nine points clear of the next-largest domain and nearly double the smallest.

It is also structurally the hardest to prepare in isolation, because it is the domain that draws on the others. A question about an agent design can simultaneously be a question about tool definitions, about what stays in the context window, and about which failure mode surfaces first. You can study Tools & MCP on its own. You cannot really study orchestration on its own, because orchestration is what the other domains get assembled into.

That is an observation about how the blueprint is constructed, not a claim about how candidates score. Anthropic does not publish per-domain pass rates, and anyone who tells you which domain candidates fail most is guessing.

Why is Claude Code a fifth of the exam?

Claude Code at 20% is the weighting that most surprises people, and it is the one most often left out of study plans built from general Claude API knowledge. A developer who has spent a year calling the Messages API in production may still have never configured Claude Code deliberately — and that is twelve questions, roughly one in five.

The gap is worth naming because it is the cheapest one to close. Unlike orchestration, which needs practice to internalise, Claude Code configuration and workflow is largely learnable from focused reading plus a few hours of deliberate use. Skipping it because it feels like tooling rather than architecture is an expensive instinct.

How should you budget study time across the domains?

Start from the weighting, then adjust for what you already do. A defensible default is to allocate hours in proportion to weight — roughly 27% of your time on agentic architecture, 20% each on Claude Code and prompt engineering, 18% on tools and MCP, 15% on context and reliability — and then apply two corrections:

1. Spend more where the exam weighting exceeds your daily exposure. If you have never used MCP, its 18% is worth more of your time than the raw number suggests. If you write production prompts every day, prompt engineering's 20% probably needs less.
2. Spend more where questions require synthesis rather than recall. Agentic architecture rewards having built something; a fact you can look up in a minute is not what a scenario question tests.

The arithmetic that matters most: at 720 out of 1000, you can miss a meaningful share of the paper and still pass, but you cannot write off an entire domain. Abandoning Context & Reliability — the smallest domain at 15% — costs you more than the margin between a comfortable pass and a fail.

What does the weighting not tell you?

Three things, all of which catch people out.

Weight is not difficulty. A 15% domain full of precise mechanical detail — caching thresholds, TTL behaviour, what happens when a prefix changes by one byte — can cost you more points per hour of neglect than a 27% domain you already work in.

Domains are not silos. The questions are scenario-based, so a single question routinely sits across two or three domains. Studying domain by domain is a good way to organise coverage and a poor way to rehearse the actual exam, which is why timed mixed practice matters closer to the date.

Per-domain question counts are approximate. The weightings are percentages of score, and the 16/12/12/11/9 split above is what those percentages imply for a 60-question paper. Treat it as a planning aid, not a guarantee about a specific exam form.

How do you turn this into a plan?

Read the official exam guide first for scope, then work the domains in weight order rather than in the order the guide lists them — you want your freshest attention on the 27%. Track your accuracy per domain rather than overall, because an overall percentage hides exactly the single-domain weakness that the weighting punishes. Finish with full-length timed practice at the real 60-question domain split, since pacing at two minutes per question is its own skill.

Plinth's practice questions, drill and mock exam are built against this same blueprint and the same 720 threshold, with the mock reproducing the official domain weighting so a practice score means something. If you are earlier in planning, our Claude Certified Architect study path covers timelines and prerequisites before you get to per-domain budgeting.