Claude Certified Architect Exam Guide

About the CCA Foundations Certification

The Claude Certified Architect (CCA) Foundations certification validates your expertise in designing and building production-grade applications with Claude. It is aimed at software engineers, solutions architects, AI engineers, and technical leads who work with Anthropic's models.

Earning the CCA credential demonstrates that you can architect reliable, secure, and efficient systems across all major Claude capabilities — from prompt engineering to agentic workflows.

Exam Format

All questions are scenario-based — you will not encounter simple definition recall. Each question presents a realistic situation requiring you to choose the best architectural decision.

Domain Breakdown

The exam is weighted across five domains. Understanding this weighting helps you allocate study time effectively:

Domain 1: Agentic Architecture & Orchestration — 27%

The largest portion of the exam. Covers autonomous agent design patterns, orchestration strategies (router, pipeline, parallel), multi-agent coordination, reliability patterns, error handling, and human-in-the-loop design.

Go deeper on Domain 1: Agentic Architecture & Orchestration — a worked scenario and the traps that catch candidates.

Domain 2: Claude Code Configuration & Workflows — 20%

Covers CLAUDE.md file structure and best practices, project-level vs user-level configuration, permission models, slash commands, and how to integrate MCP servers into Claude Code workflows.

Go deeper on Domain 2: Claude Code Configuration & Workflows — what this domain actually tests.

Domain 3: Prompt Engineering & Structured Output — 20%

System prompt design, chain-of-thought prompting, few-shot patterns, role and persona prompting, output formatting, and techniques for reducing hallucination.

Go deeper on Domain 3: Prompt Engineering & Structured Output — what this domain actually tests.

Domain 4: Tool Design & MCP Integration — 18%

Model Context Protocol fundamentals — tool interface design, structured error responses, tool distribution and tool_choice, MCP server scoping, MCP resources, and built-in tools.

Go deeper on Domain 4: Tool Design & MCP Integration — what this domain actually covers.

Domain 5: Context Management & Reliability — 15%

Token budget optimization, prompt caching (ephemeral and persistent), multi-turn conversation design, context window strategies, and managing long conversations.

Go deeper on Domain 5: Context Management & Reliability — what this domain actually tests.

The Exam Scenarios

Each CCA Foundations exam is built around six production scenarios, of which four are randomly selected for your sitting. Rather than abstract questions, every item is set inside one of these real deployment contexts — so you encounter familiar architectural problems under real-world constraints.

Customer Support Resolution Agent

An autonomous agent that handles customer enquiries end-to-end — classifying intent, retrieving relevant knowledge, drafting resolutions, and escalating to humans when confidence is low. Tests routing logic, human-in-the-loop design, and graceful degradation patterns.

Code Generation with Claude Code

A developer-facing assistant that reads codebases, proposes changes, and runs commands via Claude Code. Tests CLAUDE.md configuration, permission scoping, and the boundary between autonomous action and human approval.

Multi-Agent Research System

An orchestrator that dispatches specialised subagents to gather, synthesise, and cite information from multiple sources in parallel. Tests orchestration patterns (router, parallel, pipeline), error propagation, and output aggregation.

Developer Productivity with Claude

A team-facing tool that accelerates engineering workflows — code review, documentation generation, and onboarding assistance. Tests context management across long sessions, tool scoping, and MCP server integration.

Claude Code for Continuous Integration

Integration of Claude Code into a CI/CD pipeline to automate test writing, bug identification, and release notes. Tests agentic reliability under time constraints, failure handling, and safe automation at scale.

Structured Data Extraction

An extraction pipeline that parses unstructured documents into validated schemas for legal, financial, or operational data. Tests prompt engineering for structured output, schema design, and downstream validation patterns.

Study Strategy

Here is an effective preparation approach:

1. Learn the domains — Work through all five lesson modules to build foundational understanding
2. Practice with short sessions — Use Quick Sprint mode daily to identify weak areas
3. Focus on weaknesses — Review explanations for every wrong answer, not just the correct choice
4. Simulate real conditions — Take at least two full-length timed exams before your test date

Try a Sample CCA Exam Question

Every question on the real exam is scenario-based, not definition recall. Here's a worked example from the highest-weighted domain, Agentic Architecture & Orchestration (27%):

Sample Question — Agentic Architecture & Orchestration

An orchestrator agent spawns four subagents in parallel to process different sections of a large document. One subagent encounters a permission error on a file system tool and halts. What is the correct response from the orchestrator?

• A. Reassign the failed section to one of the three subagents that completed successfully, since their file system access is evidently working
• B. Have the failed subagent write its own error into the shared document context so the other subagents can route around it
• C. Surface the partial failure to the user, return results from successful subagents, and request clarification before retrying
• D. Grant the subagent broader file system permissions and re-run it so the job finishes without interrupting the user

Correct answer: C. A permission error is not a transient failure, so retrying it unchanged does not resolve it, and the published exam guide treats escalation as the right move when an agent cannot progress on its own. Widening the subagent’s permissions has the agent grant itself a capability it was never given. Reassigning the section to a subagent that happens to be working retries an error that fails the same way. Letting the failed subagent post its error into shared context bypasses the coordinator, which in a hub-and-spoke design owns error handling and information routing.

That's one of 400 scenario-based questions across all five domains in our CCA practice questions bank — work through the rest to build the pattern recognition the real exam rewards.

Study Timeline

How much prep time you need depends on your existing familiarity with Claude's APIs and tooling. A focused, intensive week of daily practice followed by a full timed simulation is realistic if you already work with Claude regularly. If you're newer to the ecosystem, spreading that same work across two to four weeks — covering each domain in weight order before simulating — gives you more room to absorb the material.

For a detailed day-by-day plan, see our CCA exam study schedule — or use the free CCA Study Plan Generator to build a schedule weighted to your own exam date and weak domains, no signup required.

Common Mistakes to Avoid

• Focusing only on Agentic Architecture & Orchestration because it has the highest weight — every domain can make or break your score
• Memorizing tool names without understanding when and why to use them
• Ignoring Context Management & Reliability (15%) — these questions often distinguish passing from failing scores
• Not practicing under timed conditions — time pressure changes how you approach scenario questions

Test Your Readiness

Once you understand the format and domains, take a full CCA Foundations exam to evaluate your readiness.

You can also sharpen specific areas with our CCA practice questions bank of 400 scenario-based questions across all domains.

How to Register for the Official Exam

Frequently Asked Questions

How hard is the CCA Foundations exam?

Harder than most candidates expect. Every question is scenario-based — you will not encounter simple definitions or recall prompts. The exam tests whether you can choose correctly between two architecturally plausible options under real-world constraints.

How much does the official exam cost?

The standard fee is $125 USD, paid directly to Anthropic at the time of booking. Check Anthropic's official certification page ↗ for current registration details.

What score do I need to pass?

A scaled score of 720 out of 1,000. This is domain-weighted, not a raw percentage of questions answered correctly — strong performance in a high-weight domain (Agentic Architecture & Orchestration at 27%) carries more influence on your total than equal accuracy in a lighter one.

How long is the exam?

60 questions, 120 minutes — an average of two minutes per question, though complex scenario questions in the higher-weighted domains typically take more time than shorter ones.

How long is the certification valid?

The CCA Foundations certification is valid for 12 months from the date of passing. You must renew before your 12-month term expires; if it lapses, you must re-earn the certification. Check Anthropic's official certification page ↗ for renewal details.

Who should take it?

Professionals with at least six months of hands-on production experience building Claude-powered applications. If you regularly work with Claude APIs, agentic workflows, tool use, or Claude Code in a production context, you are the target candidate. It is not suited to those new to Claude or working at a tutorial level.

Can I retake the exam if I fail?

Yes. Waiting periods increase with each failed attempt: 14 days after the first, 30 days after the second, and 90 days after the third. You may attempt the exam up to four times within a rolling 12-month period, and the exam fee applies to each attempt. For retake strategy, see our CCA exam retake policy guide.