Technology πŸ‘€ KP Expert πŸ“… Oct 02, 2026 πŸ‘οΈ 208 views

How to Prepare for the Claude Certified Architect Exam: A Study Roadmap

You can pass the Claude Certified Architect exam in about 8 weeks if you study by domain weight and build real projects instead of memorizing facts. This roadmap shows you exactly what to study each week.

Anthropic launched the Claude Certified Architect – Foundations (CCA-F) exam on March 12, 2026. It was the company's first official certification. It checks whether you can design and ship production-grade systems with the Claude API, the Claude Agent SDK, Model Context Protocol (MCP), and Claude Code.

This guide is for:

  • Developers and solution architects who already build with Claude and want to prove it
  • Cloud and DevOps engineers moving into AI architecture roles
  • Consultants at partner firms who need certified staff for their company's partner tier
If you have never built anything with Claude, start with the Claude Certified Developer or Associate track first, then come back to this one.
Exam Overview

The Foundations exam has 60 scenario-based multiple-choice questions to answer in 2 hours. The passing score is 720 out of 1000. You can take it at a test centre or online with a proctor.

There are now two Architect tracks. Both are offered through the Claude Partner Network and Anthropic Partner Academy, so check your eligibility before you plan a date.

Track
Price
Who It Is For
Architect – Foundations (CCA-F)
$125
Engineers and architects building Claude systems, with 6+ months of hands-on Claude experience
Architect – Professional
$175
Architects leading enterprise-scale deployments, with years of architecture work plus Claude in production
Prices are per attempt. Prices and eligibility can change, so always confirm on Anthropic's official certification page.

The Five Foundations Domains

Your study time should follow these weights. Agentic architecture alone is more than a quarter of the exam.

Weight
Domain
What It Covers
27%
Agentic Architecture & Orchestration
Agent loops, coordinator–subagent patterns, multi-agent design
20%
Claude Code Configuration & Workflows
CLAUDE.md hierarchy, slash commands, skills, plan mode, CI/CD
20%
Prompt Engineering & Structured Output
JSON schemas, few-shot examples, validation-and-retry loops
18%
Tool Design & MCP Integration
MCP servers, .mcp.json, built-in tools, tool descriptions
15%
Context Management & Reliability
Long context, escalation, human-in-the-loop

The Six Exam Scenarios

Questions are built around realistic production situations:

#
Scenario
1
A customer support agent with MCP tools and escalation logic
2
Code generation with Claude Code using plan mode and CLAUDE.md
3
A multi-agent research system with specialized subagents
4
Developer productivity tools using Bash, Read, Write, Grep and Glob
5
Claude Code inside a CI/CD pipeline
6
Structured data extraction with JSON schema validation

The Professional exam goes further. It covers seven domains and focuses on integration, evaluation, governance, and knowing when Claude is not the right solution.

Prerequisites Before You Start

You need working coding skills and real Claude experience before day one. The exam tests judgment, and judgment comes from building things. Check yourself against this list:

  • Comfortable in Python or TypeScript, including async code and JSON handling
  • Have called the Claude Messages API, including system prompts and multi-turn conversations
  • Have used tool use (function calling) and handled tool results
  • Have installed and used Claude Code on a real repository
  • Understand REST APIs, environment variables and basic CI/CD (for example, GitHub Actions)
  • Know basic cloud architecture: queues, retries, rate limits and logging
If fewer than four of these describe you, spend two extra weeks building small projects before following the roadmap below.
The 8-Week Study Roadmap

Plan for 8 to 10 hours a week, which works out to about 1 hour on weekdays plus a longer weekend session. Each week ends with something you have built, not just something you have read.

Week
Focus
Build by the End of the Week
1
Claude API basics (foundation)
A small chat app with a system prompt
2
Prompt engineering & structured output (20%)
An extractor that returns valid JSON, with a retry on failure
3
Tool use & MCP fundamentals (18%)
A local MCP server with two tools
4
Agentic loops & single-agent design (27%)
An agent that plans, calls tools and stops correctly
5
Multi-agent orchestration (27%)
A coordinator with two specialized subagents
6
Claude Code configuration & CI/CD (20%)
A repo with CLAUDE.md, custom commands and a CI job using headless mode
7
Context management & reliability (15%)
Escalation, human review and long-document handling added to your agent
8
Revision & practice exams (all domains)
Two timed practice tests, then fix your weakest domain
Important: if you are short on time, never cut weeks 4 and 5. Agentic architecture is the largest domain and the hardest to fake.
Core Topics to Master

Focus on the decisions an architect makes, not on syntax. Most questions describe a scenario and ask for the best design choice.

1. Agentic Architecture and Orchestration (27%)

Know when a simple prompt chain is enough and when you need an agent. Understand the agent loop: the model reasons, calls a tool, reads the result, and decides whether to continue. Learn the coordinator–subagent pattern, where one agent splits up the work and gives each subagent only the context it needs. Be ready to explain stopping conditions, how to limit runaway loops, and how to pass results between agents.

2. Claude Code Configuration and Workflows (20%)

Learn the CLAUDE.md hierarchy. A project-level file is shared through version control, while a user-level file in your home directory is personal. Practise custom slash commands, skills, plan mode and permission settings.

For CI/CD, know how to run Claude Code non-interactively in headless mode with the -p flag:

ci-review.sh
claude -p "Review the changes in this pull request for bugs and security issues. Output a short summary."
Questions often include a wrong answer that runs Claude Code interactively inside a pipeline. A CI job has no human to answer prompts, so headless mode is the correct choice.

3. Prompt Engineering and Structured Output (20%)

Write clear system prompts, use XML tags to separate instructions from data, and add few-shot examples for tricky formats. For structured output, define a JSON schema, validate every response, and retry with the validation error included in the next prompt.

4. Tool Design and MCP Integration (18%)

Good tools have clear names, specific descriptions and tight input schemas. Learn how MCP servers expose tools, resources and prompts, and how .mcp.json configures servers for a project:

.mcp.json
{
  "mcpServers": {
    "orders": {
      "command": "python",
      "args": ["orders_server.py"]
    }
  }
}

Know the trade-off between a few broad tools and many narrow ones.

5. Context Management and Reliability (15%)

Understand context window limits, summarization, and retrieval for long documents. Design escalation paths so the agent hands off to a human when its confidence is low or the action is risky. Know how prompt caching cuts cost and latency when the same context is sent repeatedly.

Hands-On Projects to Build

Build one project for each exam scenario. Each project teaches the trade-offs the exam asks about far better than reading does.

Project
Exam Scenario
Domains
Support bot with an order-lookup MCP tool and human escalation
Customer support agent
Agentic, Tools & MCP, Reliability
Feature built in a sample repo using plan mode and CLAUDE.md
Code generation with Claude Code
Claude Code
Research agent with a coordinator and two subagents
Multi-agent research system
Agentic, Context
Script that uses Claude Code to search and refactor a codebase
Developer productivity
Claude Code, Tools
GitHub Action that runs Claude Code headless to review pull requests
Claude Code for CI/CD
Claude Code
Invoice extractor with JSON schema validation and retry
Structured data extraction
Prompt engineering

Keep all six in one GitHub repository. It doubles as a portfolio when you apply for architect roles.

Study Resources

Start with the official material, then add practice questions. Anthropic Academy hosts free courses that together cover all five domains.

  • Official exam guide: read it twice and note the domain weights and scenarios
  • Anthropic Academy (anthropic.skilljar.com): free courses such as Building with the Claude API, Introduction to Model Context Protocol, Claude Code in Action and Claude 101
  • Claude documentation (docs.claude.com): the reference for tool use, prompt caching, structured output and the Agent SDK
  • Anthropic Cookbook on GitHub: working code for agents, tools and retrieval
  • MCP documentation (modelcontextprotocol.io): the specification and example servers
  • Official practice exam: take it in week 8 under timed conditions
  • KP Expert practice tests: scenario-based mock exams for the Foundations and Professional tracks
One candidate found the real exam harder than both the official practice test and the training material, so don't stop at the free practice exam.
Exam-Day Tips and Common Mistakes

You have 2 minutes per question. Read each scenario once to understand the goal and a second time to pick out the constraints.

Tips That Help

  • Choose the simplest design. Pick the simplest design that meets the requirement. Over-engineered answers are often traps.
  • Note the constraints. Cost, latency, security or β€œno human available” usually decide the right answer.
  • Manage your time. Flag long questions, answer the quick ones first, and come back to the flagged ones with the remaining time.
  • Check your setup. For online proctoring, test your webcam, microphone and internet connection the day before.

Mistakes to Avoid

Mistake
Why It Costs You
Mixing up tool conventions
Cursor or Codex conventions are not the same as Claude Code's, and the exam tests Claude Code specifically
Interactive Claude Code in CI/CD
A pipeline has no human to answer prompts, so headless mode is the right answer
Relying only on AI-generated drills
Drills test recall, while the exam tests judgment you only get from building
Equal time on every domain
Agentic architecture is worth 27%, while context management is worth 15%
Over-engineering with multi-agent designs
A single agent or a prompt chain is often the correct, simpler answer
Frequently Asked Questions

How many questions are on the Claude Certified Architect – Foundations exam?

The exam has 60 multiple-choice questions, and you have 2 hours to finish.

What is the passing score?

You need 720 out of 1000.

How much does the exam cost?

The Foundations exam costs $125 per attempt and the Professional exam costs $175 per attempt. Check Anthropic's official page for current prices.

Can anyone take the exam?

Access is currently through the Claude Partner Network and Anthropic Partner Academy. If your company is not a partner, check the official certification page for current eligibility.

How long should I study?

Most working professionals need 6 to 8 weeks at 8 to 10 hours a week, provided they already have hands-on Claude experience.

Should I take Foundations or Professional first?

Take Foundations first unless you already lead enterprise Claude deployments. Professional focuses on governance, evaluation and integration at scale.

Is the exam open book?

No. It is proctored, and it tests judgment you build through practice rather than facts you can look up.

Conclusion

Passing the Claude Certified Architect exam comes down to three habits: study by domain weight, build one project per scenario, and practise under timed conditions. Follow the 8-week plan above and you will walk into the exam with real experience, not just notes.

Recap: what to take away

  • The Foundations exam has 60 questions in 2 hours, and you need 720 out of 1000 to pass
  • Spend the most time on agentic architecture (27%), then Claude Code and prompt engineering (20% each)
  • Build one hands-on project for each of the six exam scenarios
  • Use headless mode for CI/CD, and choose the simplest design that meets the requirement
  • Finish with timed practice exams and fix your weakest domain

Ready to start? KP Expert offers scenario-based prep courses for the Claude Certified Architect – Foundations and Professional exams, with practice tests mapped to every domain. Explore the courses at KP Expert.

CLAUDE CERTIFICATION SERIES Published by KP Expert β†’
Recently Enrolled

Student enrolled in this course.

View course
Explore Courses

Latest from @kp__expert

Follow on Instagram
Loading Instagram posts...

AI Course Assistant

Share your details and goals to get the best course recommendations.

Recommended Courses

Select a course name to view full details.

Course Details
Enrollment & Contact
  • Review selected course and confirm your enrollment request.
  • Click checkout to move into the full payment process.
  • After payment submission, your enrollment is processed by our team.
Admissions Contact
Email: info@kpexpert.com
Phone: +91 92708 37105
Your submitted details
Name, email and phone will appear here.
Your request has been submitted successfully. Our team will contact you shortly.