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 5 official domains: Agentic Architecture, Claude Code, Prompt Engineering & Structured Output, Tools & MCP, and Context Management & Reliability.

How do I prepare for the CCA-F exam?

This free study platform offers 116 lessons across the 5 official exam domains, 800+ practice questions, unlimited weighted mock exams, and spaced-repetition flashcards to prepare for the CCA-F exam. It features interactive quizzes, diagnostics, and progress tracking.

What are the 5 official CCA-F exam domains?

The CCA-F exam covers 5 official domains weighted as follows: Agentic Architecture & Orchestration (27%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%), Tool Design & MCP Integration (18%), and Context Management & Reliability (15%). A solid background in Claude fundamentals is assumed.

How many questions are on the CCA-F exam?

The CCAR-F (Claude Certified Architect - Foundations, formerly CCA-F) exam has 60 items in 120 minutes, mixing multiple-choice and multiple-response questions. It is scored on a 1000-point scale with 720 required to pass. It is proctored (online via Pearson OnVUE or at a Pearson test center), closed-book, and valid for 12 months, renewed via a free non-proctored assessment on the Anthropic Partner Academy.

How long does the CCA-F exam take?

The CCA-F exam allows 120 minutes (2 hours) to complete 60 questions. Candidates average roughly 2 minutes per question. Good time management means reading each scenario carefully and flagging uncertain answers for review rather than lingering.

What is the passing score for the CCA exam?

The CCA-F exam requires a passing score of 720 out of 1000 points. The exam mixes multiple-choice and multiple-response items, is closed-book and proctored (Pearson OnVUE online or a Pearson test center). It is designed to be difficult, only well-prepared candidates should expect to pass.

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. The stop_reason field signals when the task is done. Agentic architecture accounts 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 5 official domains weighted as follows: Agentic Architecture & Orchestration (27%), Claude Code (20%), Prompt Engineering & Structured Output (20%), Tools & MCP (18%), and Context Management & Reliability (15%).

Can I take the CCA practice exam for free?

Yes. This platform provides unlimited free mock exams (60 questions, 120 minutes, weighted to the official domain split), domain-specific practice, and flashcards. No registration or payment is required.

How do I get started with Claude AI development?

Start with the Fundamentals: 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), and Haiku (fast and cost-effective for simple tasks).

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 5 official domains: Agentic Architecture, Claude Code, Prompt Engineering & Structured Output, Tools & MCP, and Context Management & Reliability.

How do I prepare for the CCA-F exam?

This free study platform offers 116 lessons across the 5 official exam domains, 800+ practice questions, unlimited weighted mock exams, and spaced-repetition flashcards. All content is free and open-source.

What are the 5 official CCA-F exam domains?

The exam covers 5 official domains weighted as follows: Agentic Architecture & Orchestration (27%), Claude Code Configuration & Workflows (20%), Prompt Engineering & Structured Output (20%), Tool Design & MCP Integration (18%), and Context Management & Reliability (15%).

How many questions are on the CCA-F exam?

The CCAR-F (formerly CCA-F) exam has 60 items mixing multiple-choice and multiple-response questions with a 120-minute time limit. The passing score is 720 out of 1000. The exam is proctored (Pearson OnVUE online or a Pearson test center) and closed-book.

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 116 lessons and pass the certification in 4 to 8 weeks with consistent study. The platform tracks your progress across all 5 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 CCAR-F (Claude Certified Architect - Foundations) exam is 720 out of 1000 possible points. The exam has 60 items mixing multiple-choice and multiple-response questions and a 120-minute time limit. Credentials are valid 12 months with free non-proctored renewal on the Partner Academy.

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 unlimited mocks weighted to the official 27/20/20/18/15 split, 60 questions, 120 minutes, and 720/1000 scoring. Each exam includes detailed explanations 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. It provides agent lifecycle management, state persistence, tool binding, inter-agent communication, and observability.

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, adaptive thinking), and Haiku for simple high-throughput tasks (fastest, lowest cost). All share the same Messages API.

All Lessons

Under Amdahl's Law, why is spending engineering sprints optimizing vector search latency from 50 ms down to 5 ms considered a 'micro-optimization trap' in an end-to-end RAG system?

AVector search engines automatically revert latency optimizations after 24 hours.BReducing vector search latency causes cross-encoders to emit lower relevance scores.CVector databases charge higher API fees for queries that complete in under 10 milliseconds.DLLM generation accounts for 80 to 90 percent of total latency (e.g. 1,500 ms); saving 45 ms in retrieval yields less than a 3 percent improvement in overall response time.

1 / 1

advanced 45 min

Day 27: Optimize Latency and Cost by Stage

Optimize Latency and Cost by Stage

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

Loading lesson content...

In PolicyDesk, how does ordering prompt context from lowest volatility (static instructions and core manual) to highest volatility (user role and query) as shown above optimize performance?

AIt prevents the model from generating Markdown citation links in its response.BIt maximizes byte-level prefix matching for Anthropic prompt caching, cutting input token pricing by 90 percent and reducing time-to-first-token.CIt forces the language model to ignore all user-provided query text.DIt converts the user query into a vector embedding inside the language model.

1 / 1

FieldDesk technicians operating over remote satellite links experience 7-second response times because the orchestrator feeds 35 unranked candidate chunks (14,000 tokens) into a flagship model. Which stage-level optimization pipeline achieves sub-2-second responses without degrading safety-critical accuracy?

ABypass retrieval entirely and allow the model to answer from pre-trained parametric memory.BDownscale technician tablet displays to 480p resolution to reduce bandwidth consumption.CRetrieve 50 candidates via fast hybrid search, use a cross-encoder to prune down to the top 5 relevant chunks (under 1,800 tokens), and route simple lookups to Claude 3.5 Haiku while reserving Claude 3.5 Sonnet for complex synthesis.DIncrease the number of retrieved chunks from 35 to 70 to maximize context redundancy.

1 / 1

Warm-UpQuick check before we dive in

Day 27: Optimize Latency and Cost by Stage

On this page

Check Your UnderstandingTest your knowledge so far

Lesson ReviewSolidify what you've learned