Claude Certified Architect – Foundations: Complete Exam Guide
Claude Certified Architect – Foundations is Anthropic’s architect-level foundations credential for designing Claude-based solutions. Anthropic’s current Academy page lists a 60-question, 120-minute exam priced at $125 USD, a 12-month credential validity period, a passing score of 720 on a 100–1,000 scaled range, and delivery through online proctoring or Pearson test centers. The public credential page presents the full certification name and five weighted domains; PrepAway’s source inventory uses CCA-F as its internal exam identifier, but this guide uses the official credential name for factual claims.
On this page
- Current credential and availability
- Exam format and logistics
- The five official domains
- Agentic Architecture & Orchestration
- Tool Design & MCP Integration
- Claude Code Configuration & Workflows
- Prompt Engineering & Structured Output
- Context Management & Reliability
- Build a study strategy by weight
- Turn domains into practical labs
- Judge readiness with evidence
- Scheduling and credential lifecycle
- Final preparation checklist
Current credential and availability
Anthropic currently lists Claude Certified Architect – Foundations in its Partner Academy certification program. The credential is aimed at people designing Claude-based systems rather than only writing individual prompts or consuming one product feature.
The Academy material states that certification exams are available to members of the Claude Partner Network. That makes eligibility different from a broadly open public certification program, so verify your current access before building a study schedule around a target date.
Anthropic launched Architect – Foundations in March 2026 and later expanded its certification catalog with additional role-based credentials. That matters because Claude development, enterprise operations, and architect-level work now have separate certification intents.
PrepAway maps this exam to the Anthropic certification ecosystem and the existing CCA-F exam page.
Exam format and logistics
Anthropic’s current credential page lists 60 questions and 120 minutes of exam time. Academy notes also describe approximately 135 minutes of seat time when administrative steps are included.
The listed price is $125 USD. The exam is delivered in English and can be taken through online proctoring or at a Pearson test center, according to the current Academy information.
The reported passing score is 720 on a scaled score range from 100 to 1,000. A scaled score is not the same as a simple percentage, so avoid trying to convert 720 into a fixed raw-question target.
The credential is valid for 12 months. Treat certification maintenance as part of the planning decision because Claude products, agent patterns, and developer-platform behavior can change quickly.
The five official domains
Anthropic publishes five weighted domains for Claude Certified Architect – Foundations: Agentic Architecture & Orchestration at 27%; Tool Design & MCP Integration at 18%; Claude Code Configuration & Workflows at 20%; Prompt Engineering & Structured Output at 20%; and Context Management & Reliability at 15%.
The weights show that this is not a prompt-writing exam with a few architecture questions added. The largest share belongs to agentic architecture and orchestration, while tools, Claude Code, prompting, and context management together make up the rest of the blueprint.
Study by weight, but do not isolate the domains. A production agent can combine orchestration, tool schemas, MCP connectivity, Claude Code configuration, structured outputs, context strategy, evaluation, and human approval in one design.
The Claude Production Engineering pillar is the strongest high-level companion to this cluster, while Claude Development covers tool, prompt, code, and integration practices.
Domain 1: Agentic Architecture & Orchestration — 27%
The largest domain is about choosing and designing the right coordination pattern for a problem. You should be comfortable distinguishing a simple prompt-response call from a deterministic workflow and from a more autonomous agent loop.
Anthropic’s engineering guidance recommends starting with the simplest architecture that solves the task and increasing autonomy only when the extra flexibility produces measurable value. That principle helps with exam scenarios because complexity is a cost as well as a capability.
Study sequential workflows, parallel work, routing, evaluator-optimizer patterns, single-agent loops, multi-agent delegation, stop conditions, retries, human approval boundaries, and state handoff between steps.
Use Agentic Architecture and Orchestration as the domain overview and connect it to production failure handling rather than memorizing pattern names alone.
Domain 2: Tool Design & MCP Integration — 18%
Tool design is partly interface design for a model. Names, descriptions, schemas, response shape, error behavior, authorization, and context efficiency all influence whether a tool is selected and used correctly.
Model Context Protocol adds a standardized way to connect AI applications with external systems. MCP servers can expose capabilities such as tools, resources, and prompts, while clients decide how those capabilities are discovered and used.
Prepare to reason about boundaries: which operation should be a tool, what data should be exposed as a resource, where authentication belongs, how much context a tool response should return, and when human approval is required.
The cluster separates tool design from the dedicated MCP architecture guide so neither article becomes an unfocused catalog.
Domain 3: Claude Code Configuration & Workflows — 20%
Claude Code is treated as an operational development environment, not only a chat interface. Preparation should include how project instructions, permission settings, repository context, external tools, and reusable workflow rules shape behavior.
Current Anthropic guidance emphasizes keeping project instructions concise, organizing reusable or scoped rules rather than building one giant instruction file, and controlling permissions according to the consequence of an action.
Study how Claude Code learns a repository, how project-level guidance differs from one-off prompts, how MCP integrations add external systems, how planning and approval modes change execution, and how teams keep automation reviewable.
The Claude Code Configuration and Workflows article is the direct Domain 3 study anchor.
Domain 4: Prompt Engineering & Structured Output — 20%
Prompt engineering at architect level is about producing reliable application behavior, not finding one magic phrase. System instructions, examples, task decomposition, data boundaries, tool descriptions, output contracts, and evaluation all matter.
Structured outputs provide schema-constrained JSON responses and strict tool-use schemas for machine-facing interfaces. The schema can guarantee structure, but it does not replace business validation, authorization, or semantic evaluation.
Prepare to distinguish natural-language guidance from machine contracts. A schema can require a field to be an integer while the application still has to decide whether the integer is allowed or sensible.
The rewritten Prompt Engineering and Structured Outputs guide is the Domain 4 survivor and will absorb useful JSON material from the overlapping article later.
Domain 5: Context Management & Reliability — 15%
Context should be treated as a limited architectural resource. More text is not automatically better if low-value history, stale tool results, duplicate instructions, or irrelevant documents reduce the signal available to the model.
Study context windows, retrieval, memory, compaction, prompt caching, evaluation, guardrails, latency, and recovery behavior as connected reliability choices.
Reliable systems also need observability and evaluation. A response can be syntactically valid and still fail the product objective, so architecture should define what success means and measure it with representative cases.
The cluster will broaden Context Management and Reliability into the official Domain 5 overview, supported by memory, caching, retrieval, evaluation, guardrail, and latency articles.
Build a study strategy from domain weight and current skill
Start with a diagnostic rather than dividing study time equally. If you want a compressed schedule after that diagnostic, use the 30-Day CCA-F Study Plan to sequence the five domains without giving every topic equal time. An experienced application architect may already understand orchestration and reliability but need more Claude Code configuration detail. A developer using Claude Code daily may need deeper agent architecture and MCP design.
A simple weighting model is useful: give the 27% architecture domain the largest share, then treat the two 20% domains as major study blocks, followed by tools/MCP at 18% and context/reliability at 15%.
Within each block, alternate reading with explanation. Describe why you would choose a workflow instead of an agent, how you would constrain a high-impact tool, how project instructions should be organized, or why a schema does not replace business validation.
Keep one error and uncertainty log. If you cannot explain a design decision without looking at notes, the objective is not finished.
Turn the blueprint into practical labs
For architecture, build a small workflow that routes between a deterministic path and an agentic path, then define stop conditions and an approval boundary.
For tool and MCP work, design a small capability with a narrow schema, concise response, explicit errors, and least-privilege access. Then document what information should be a tool versus a resource.
For Claude Code, create a safe practice repository with concise project instructions, scoped rules, permission choices, and one external integration. Observe how configuration changes affect the workflow.
For structured output and context, define a schema-backed response, test invalid business values separately, and run representative evaluations under different context lengths or retrieval choices.
Judge readiness with evidence instead of course completion
You are approaching readiness when you can map an unfamiliar scenario to one or more official domains and explain which architecture decision matters most.
You should be able to compare workflow patterns, design a tool boundary, explain an MCP integration, describe Claude Code configuration choices, write a prompt/output contract, and diagnose a context or reliability failure.
Practice should include tradeoffs. The best design is rarely the one with the most agents, the largest context, or the greatest number of tools.
Use current official material near exam day because Claude platform and Claude Code behavior can evolve faster than a traditional infrastructure certification.
Scheduling and credential lifecycle
Before scheduling, confirm that your Partner Network access is active, the Academy page still lists the same exam logistics, and your chosen delivery method meets any current proctoring requirements.
Plan enough time for the five domains and for practical work. A short schedule is reasonable only when your diagnostic shows existing experience across Claude architecture, tools, Claude Code, prompts, and reliability.
Because the credential is listed as valid for 12 months, certification value comes partly from staying current with Anthropic’s platform after the exam.
Do not over-focus on an exam-code label. The current public credential page is authoritative for the certification name and blueprint; prepare to that official scope.
Final preparation checklist
Know the five domains and weights, understand the current delivery and validity details, and confirm your access through the Claude Partner Network.
Practice architecture choices, tool and MCP boundaries, Claude Code configuration, prompt/output contracts, and context/reliability decisions with small real examples.
Use current Anthropic guidance for fast-moving product behavior and use PrepAway’s CCA-F cluster to connect the official blueprint to focused practice topics.
The goal is not to memorize Claude terminology. It is to show that you can design a reliable Claude solution and explain why its architecture is appropriate.