What is the Claude Certified Architect certification?

The Claude Certified Architect (CCA) certification, offered by Anthropic, validates expertise in designing, building, and deploying AI applications using Claude. The Foundations (CCA-F) level covers prompt engineering, context engineering, tool use, agent design, Model Context Protocol (MCP), Claude Code, and production architecture across 12 domains.

How do I prepare for the CCA-F exam?

This free study platform offers 72 structured lessons across 12 domains, 715 practice questions, domain-specific exams, and 10 full certification simulation variants to prepare for the CCA-F exam. It features spaced repetition, interactive quizzes, and knowledge mapping.

What is covered on the CCA-F exam?

The CCA-F exam covers 12 domains: Foundations and Core Concepts, Prompt Engineering, Context Engineering, Tool Use and Function Calling, Structured Outputs, Model Context Protocol (MCP), Claude Code, Claude Agents SDK, Agentic Architecture, Production Architecture, Reliability Engineering, and Multi-Agent Systems.

How many questions are on the CCA-F exam?

According to Anthropic, the CCA-F exam contains 60 questions to be completed within 120 minutes. The passing score is 720 out of 1000. Question types include multiple choice (30%), multiple select (20%), scenario-based (25%), architecture review (15%), and tradeoff analysis (10%).

How long does the CCA-F exam take?

The CCA-F exam allows 120 minutes (2 hours) to complete 60 questions. With 25% scenario-based questions, time management is critical. Candidates have approximately 2 minutes per question and should allocate more time to multi-step scenario and tradeoff analysis questions.

What is the passing score for the CCA exam?

The CCA-F exam requires a passing score of 720 out of 1000 points. The difficulty distribution is approximately 25% easy questions, 50% medium questions, and 25% hard questions. Domain practice exams on this platform require 70% to pass, while full certification simulations require 75%.

Is the Claude Certified Architect certification worth it?

The Claude Certified Architect certification demonstrates verified expertise in building AI applications with Claude. As organizations increasingly adopt AI agents and Claude-powered workflows, certified architects are positioned to lead architecture decisions, optimize production deployments, and implement best practices across multi-agent systems.

Claude Code is a CLI tool developed by Anthropic that brings Claude's coding capabilities directly into the development environment. It supports edit mode and plan mode, keyboard shortcuts, slash commands, and a skills directory. Claude Code accounts for approximately 20% of the CCA-F exam content.

How do Claude agents work?

Claude agents operate using an agentic loop: send a request to the Messages API, process Claude's response, execute any tool calls Claude requests, append the results, and repeat until Claude returns end_turn. The stop_reason field is the sole termination signal. Agentic architecture and the Agents SDK together account for approximately 27% of the CCA-F exam.

What is the Model Context Protocol (MCP)?

The Model Context Protocol (MCP) is an open standard developed by Anthropic and governed by the Linux Foundation's Agentic AI Foundation. It provides a three-layer architecture (Host, Client, Server) with three primitives: Resources (application-controlled), Tools (model-controlled), and Prompts (user-controlled), built on JSON-RPC 2.0.

What domains are tested on the CCA exam?

The CCA-F exam tests 12 domains: Foundations (2%), Prompt Engineering (12%), Context Engineering (8%), Tool Use and Function Calling (9%), Structured Outputs (8%), MCP (9%), Claude Code (20%), Agents SDK (12%), Agentic Architecture (15%), Production Architecture (0%), Reliability Engineering (5%), and Multi-Agent Systems (0%). The highest-weight domains are Agentic Architecture, Claude Code, and Prompt Engineering.

Can I take the CCA practice exam for free?

Yes. This platform provides free domain practice exams (15 questions, 20 minutes each) and 10 full certification simulation variants (65 questions, 130 minutes each). All practice tests include detailed explanations and track performance across attempts. No registration or payment is required.

How do I get started with Claude AI development?

Start with the Foundations domain: learn what Claude is, understand the Messages API, manage tokens, write system prompts, and configure sampling parameters. Claude offers three model tiers: Sonnet (general-purpose), Opus (complex reasoning with extended thinking), and Haiku (fast and cost-effective for simple tasks).

What is the Claude Certified Architect certification?

The Claude Certified Architect (CCA) certification validates expertise in designing, building, and deploying AI applications using Claude by Anthropic. It covers prompt engineering, context engineering, tool use, agents, and production architecture across 12 domains.

How do I prepare for the CCA-F exam?

This free study platform offers 72 structured lessons across 12 domains, 715 practice questions, 12 domain exams, and 10 full certification simulations to prepare for the CCA-F exam. All content is free and open-source.

What topics does the Claude Certified Architect exam cover?

The exam covers 12 domains: Foundations and Core Concepts, Prompt Engineering, Context Engineering, Tool Use and Function Calling, Structured Outputs, Model Context Protocol (MCP), Claude Code, Agents SDK, Agentic Architecture, Production Architecture, Reliability Engineering, and Multi-Agent Systems.

How many questions are on the CCA-F exam?

The CCA-F exam has 60 questions with a 120-minute time limit. The passing score is 720 out of 1000. Question types include multiple choice, multiple select, scenario-based, architecture review, and tradeoff analysis.

Is the CCA-F exam free?

The study platform is completely free and open-source. The CCA-F certification exam itself is administered by Anthropic and requires payment. This platform provides all the preparation materials you need at no cost.

How long does it take to get Claude Certified Architect certified?

Most learners complete the 72 lessons and pass the certification in 4 to 8 weeks with consistent study. The platform tracks your progress across all 12 domains with spaced repetition and learning analytics to optimize your study time.

What is the passing score for the CCA-F exam?

The passing score for the CCA-F exam is 720 out of 1000 possible points. The exam has 60 questions and a 120-minute time limit. Question types include scenario-based, architecture review, and multiple choice formats.

What is Model Context Protocol (MCP)?

Model Context Protocol (MCP) is an open protocol developed by Anthropic that standardizes how AI applications connect to external systems. It uses a three-layer architecture (Host, Client, Server) with three primitives: Resources, Tools, and Prompts.

How do practice exams help with the CCA-F certification?

Practice exams simulate the real CCA-F testing experience with 12 domain-specific exams (15 questions each, 20 minutes) and 10 full certification variants (65 questions, 130 minutes). Each exam includes detailed scoring and performance analytics.

What is Claude Code and how is it used?

Claude Code is a CLI tool that brings Claude's coding capabilities directly into your terminal with file system access, command execution, and project context. It features edit mode for making changes and plan mode for safe exploration and analysis.

What is the Agents SDK in Claude?

The Claude Agents SDK is a framework for building stateful, long-running AI agents on Cloudflare Workers. It provides agent lifecycle management, state persistence, tool binding, inter-agent communication, and observability, available in both Python and TypeScript.

What are the three Claude model tiers?

Claude offers three model tiers: Sonnet for general-purpose tasks (fast, high intelligence, moderate cost), Opus for complex reasoning (highest intelligence, slower), and Haiku for simple high-throughput tasks (fastest, lowest cost). All share the same Messages API.

All Lessons

A data analytics company is building an AI agent that generates SQL queries from natural language. Each request includes: database schema context (40K), 3 few-shot examples (15K), the user's question (2K), and metadata (1K). Total budget is 200K. The architect wants to increase examples to 10 to cover more patterns. An engineer advises against this, arguing the largest share should go to a different section. Which section should receive the largest allocation?

ASystem instructions — they define the database schema and query patterns, which are the most important information for generating correct SQLBUser input and task-specific context — this is what varies per request and directly determines the output qualityCFew-shot examples — more examples always improve accuracy, and the budget should accommodate as many as possibleDConversation metadata — tracking user preferences and history is critical for personalized query generation

1 / 3

intermediate 20 min

Token Budgeting

Based on Anthropic Claude documentation and the CCA-F exam blueprint. Updated June 2026.

Loading lesson content...

An edtech platform is building a study assistant that needs to reference the full content of 20 textbooks (approximately 5M tokens total). The 200K context window is far too small. The team considers upgrading to an enterprise plan, compressing with gzip, or splitting into 25 parallel sub-requests. Which approach correctly solves the context window limitation?

AUpgrade to an enterprise plan — higher-tier plans offer larger context windows and priority accessBImplement RAG — embed textbook chapters into a vector database and retrieve only the 3-5 most relevant chunks per student questionCCompress the textbook text using gzip before sending — compression reduces the byte size and the API will count it as fewer tokensDSplit the question into 25 parallel requests, each analyzing one textbook, then merge results client-side

1 / 1

A customer support startup is evaluating token cost over a 50-turn conversation. Raw history totals 400K tokens, but the team uses hierarchical summarization. After 50 turns, each request includes: a 5K system prompt, the current user message (2K), and a summary of past turns (15K). The CTO wants the total cost. What is the correct way to calculate the effective token cost?

ASum all tokens from all 50 requests — this captures the total infrastructure cost of running the conversationBCalculate the cost per individual request based on the summarized prompt (5K + 15K + 2K = 22K tokens) and multiply by the number of requestsCUse the average token cost per message across all 50 turns as the standard cost metricDAssume a fixed cost per conversation since summarization keeps the per-request cost roughly constant

1 / 1

Warm-UpQuick check before we dive in

Token Budgeting

On this page

Check Your UnderstandingTest your knowledge so far

Lesson ReviewSolidify what you've learned