Claude Certified Developer – Foundations: Important Topics, Notes and Practice Questions

Claude Certified Developer – Foundations

Important Topics, Concept Notes, Real-Life Examples and Practice Questions

Preparation note: These are original exam-style practice questions for learning. They are not official Anthropic exam questions or leaked exam content.

Preparing for the Claude Certified Developer – Foundations exam requires understanding how Claude applications are designed, integrated, secured, tested, and operated. Each topic below first explains the concept, then gives a real-life example, followed by practice questions with answers and explanations.

Table of Contents

1. 1. Claude API, SDK and Application Integration
2. 2. Prompt Engineering and Context Engineering
3. 3. Structured Output, Tool Validation and Reliability
4. 4. Agents, Subagents and Multi-Step Workflows
5. 5. Tools, Skills, MCP and Claude Code
6. 6. Security, Privacy and Guardrails
7. 7. Model Selection, Cost, Evaluation and Production Operations

1. Claude API, SDK and Application Integration

Topic Explanation

This topic explains how developers connect applications to Claude using the API and SDK. Focus on asynchronous calls, client usage, tool execution, request handling, and error management.

Important Concepts to Remember

• Claude API and SDK clients
• Synchronous vs asynchronous calls
• Tool-use request and response flow
• Retries and transient errors
• Separating application logic from model calls

Real-Life Example

A customer-support REST service receives many requests at the same time. An asynchronous Claude SDK client allows the service to call Claude without blocking other requests.

Example Exam-Style Questions

Question 1: Your team is building an asynchronous REST service that calls Claude. Which approach is most appropriate?

1. Use the asynchronous Claude SDK client
2. Use only synchronous calls
3. Create a new client for every request
4. Avoid the SDK entirely

Correct Answer: A
An asynchronous client fits an asynchronous service and avoids blocking request handling.

Question 2: What is the main purpose of a clear tool schema?

1. To increase temperature
2. To help Claude produce valid tool arguments
3. To remove the need for validation
4. To increase token usage

Correct Answer: B
A schema describes the expected arguments and supports reliable validation.

Question 3: Which language is commonly used with the Claude SDK?

1. Python
2. Only HTML
3. Only SQL
4. Only CSS

Correct Answer: A
The Claude SDK is commonly used from programming languages such as Python and TypeScript.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

2. Prompt Engineering and Context Engineering

Topic Explanation

This topic covers how to write reliable instructions and manage the information supplied to Claude. The goal is to keep instructions clear, relevant, reusable, and within the available context.

Important Concepts to Remember

• System prompts
• Clear task instructions
• Prompt version control
• Context windows
• Prompt caching
• Context pruning and summarization

Real-Life Example

A research assistant receives a long conversation containing outdated search results. The application summarizes old results and keeps only the active research requirements in the current context.

Example Exam-Style Questions

Question 1: Where should common application-wide instructions normally be placed?

1. In every user message
2. In the system prompt
3. Inside database records
4. Inside tool output

Correct Answer: B
System instructions provide stable, high-level behavior guidance.

Question 2: A long conversation contains old tool results that are no longer relevant. What should you do?

1. Keep every result forever
2. Increase temperature
3. Prune or summarize older context
4. Restart the operating system

Correct Answer: C
Context engineering keeps the active context focused on information needed for the current task.

Question 3: An application repeatedly sends the same long instructions. Which feature can reduce repeated input processing?

1. Prompt caching
2. Higher temperature
3. More tools
4. Random sampling

Correct Answer: A
Prompt caching can reuse stable prompt portions when the request pattern supports it.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

3. Structured Output, Tool Validation and Reliability

Topic Explanation

This topic explains how to make Claude responses usable by software. Developers should request the expected format, validate model output, and check tool arguments before execution.

Important Concepts to Remember

• JSON and structured output
• JSON schema validation
• Tool argument validation
• Malformed response handling
• Output checks before downstream processing

Real-Life Example

An invoice application asks Claude to return invoice details as JSON. Before saving the result, the application checks that invoice number, amount, currency, and customer fields match the required schema.

Example Exam-Style Questions

Question 1: Claude returns malformed JSON. What is the most appropriate response?

1. Ignore the error
2. Validate the output against a schema
3. Send it directly to the database
4. Increase temperature

Correct Answer: B
Structured output should be checked before downstream code uses it.

Question 2: How should tool arguments be handled before execution?

1. Execute them without checking
2. Validate them against the tool schema
3. Convert all values to strings
4. Remove optional fields automatically

Correct Answer: B
Validation prevents malformed or unsafe arguments from reaching the tool.

Question 3: Your application requires a specific JSON structure. Which approach is most reliable?

1. Request structured output and validate it
2. Trust the model completely
3. Use free-form text only
4. Remove output restrictions

Correct Answer: A
The prompt defines the desired structure, while validation verifies that the response complies.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

4. Agents, Subagents and Multi-Step Workflows

Topic Explanation

This topic covers agent architecture for tasks that require planning, tool calls, multiple steps, or specialized reasoning. An orchestrator can delegate work to focused subagents and combine the results.

Important Concepts to Remember

• Orchestrators
• Specialized subagents
• Multi-step workflows
• Checkpointing
• Resume support
• Separating triage from response generation

Real-Life Example

A support ticket contains billing, technical, and account-access problems. An orchestrator sends each issue to a specialized subagent and then combines the findings into one response.

Example Exam-Style Questions

Question 1: A support ticket contains billing, technical and account-access issues. Which architecture is suitable?

1. One unstructured prompt
2. An orchestrator with specialized subagents
3. Three unrelated databases
4. A single static response

Correct Answer: B
An orchestrator can route separate issues to specialized subagents and combine their results.

Question 2: A long-running agent workflow stops midway. What should be used to resume it safely?

1. Checkpointing
2. Higher temperature
3. Deleting the conversation
4. Repeating every step from the beginning

Correct Answer: A
Checkpointing stores progress so a workflow can resume without repeating completed work.

Question 3: Why might ticket triage and response generation use separate Claude calls?

1. To separate classification from response creation
2. To remove validation
3. To increase hallucinations
4. To avoid tools

Correct Answer: A
Separating stages makes each task easier to validate and troubleshoot.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

5. Tools, Skills, MCP and Claude Code

Topic Explanation

This topic explains how Claude connects to external capabilities. Tools expose actions, Skills provide reusable capabilities, MCP standardizes access to tools and resources, and CLAUDE.md supplies project instructions in Claude Code workflows.

Important Concepts to Remember

• Tool definitions and descriptions
• Skills and reusable capabilities
• MCP servers
• Resources, tools, and prompts
• CLAUDE.md instructions
• Project and directory-level guidance

Real-Life Example

A development team uses an MCP server to expose approved database-search and documentation tools. Claude Code also reads project instructions from CLAUDE.md before modifying files.

Example Exam-Style Questions

Question 1: Which description best fits an MCP server?

1. A service exposing tools, resources or prompts to compatible clients
2. A database backup file
3. A CSS framework
4. A model temperature setting

Correct Answer: A
MCP standardizes how compatible AI clients access external capabilities and context.

Question 2: When is a reusable Skill or MCP capability useful?

1. When a capability must be shared across workflows or teams
2. Only for changing font colors
3. Only for storing passwords in source code
4. Never in production

Correct Answer: A
Reusable capabilities reduce duplication and provide a consistent interface.

Question 3: What is the purpose of CLAUDE.md in Claude Code workflows?

1. To provide project or directory-level instructions
2. To store API keys in plain text
3. To replace all source code
4. To increase model temperature

Correct Answer: A
CLAUDE.md provides instructions and project context that Claude Code can use.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

6. Security, Privacy and Guardrails

Topic Explanation

This topic focuses on protecting sensitive information and preventing unsafe actions. Treat external input as untrusted data, keep secrets outside source code, validate actions, and add approval controls for risky operations.

Important Concepts to Remember

• Sensitive-data redaction
• Application-boundary protection
• Untrusted input
• Prompt-injection resistance
• Secrets managers and environment configuration
• Human approval and deterministic guardrails

Real-Life Example

A finance assistant can prepare a payment but cannot submit it automatically. The application validates the payment details and requires an approved human action before execution.

Example Exam-Style Questions

Question 1: Your application receives sensitive customer information. Where should redaction occur?

1. Only after the model responds
2. At the application boundary before sending data
3. Inside the final blog post
4. Nowhere

Correct Answer: B
Sensitive information should be minimized or redacted before it is sent to the model.

Question 2: How should untrusted user input be treated?

1. As trusted system instructions
2. As data that must be validated and sanitized
3. As executable code
4. As administrator commands

Correct Answer: B
User input should not automatically gain instruction-level authority.

Question 3: Before Claude can delete database records, what is appropriate?

1. Execute automatically
2. Use deterministic checks and approval controls
3. Increase temperature
4. Remove the tool schema

Correct Answer: B
Destructive operations need safeguards such as validation, permissions and human approval where appropriate.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

7. Model Selection, Cost, Evaluation and Production Operations

Topic Explanation

This topic covers choosing models based on quality, latency, and cost; monitoring production behavior; testing model changes; and measuring usage. The correct model is determined by the application's requirements, not simply by model size.

Important Concepts to Remember

• Quality, latency, and cost targets
• Input and output token tracking
• Batch processing
• Smaller models for simple tasks
• Production traces and logs
• Evaluation tests, feature flags, and rollback

Real-Life Example

A company uses a smaller, faster model for high-volume ticket classification and a more capable model for complex investigations. It tracks token usage by feature and tests model upgrades before full rollout.

Example Exam-Style Questions

Question 1: What should be clarified before selecting a model?

1. Quality, latency and cost requirements
2. Only the model name
3. Only the prompt length
4. Only the interface color

Correct Answer: A
Model selection should match the application's quality, speed and budget requirements.

Question 2: Which option suits 10,000 flexible-schedule tasks processed overnight?

1. Batch API
2. Manual processing only
3. Browser automation
4. Repeated synchronous retries

Correct Answer: A
Batch processing is appropriate when work can run asynchronously within a flexible time window.

Question 3: What should be tracked to understand AI application cost?

1. Input tokens, output tokens, model usage and feature volume
2. Only number of users
3. Only response length
4. Only server uptime

Correct Answer: A
Cost analysis needs usage data connected to models, tokens and product features.

Question 4: A production response is incorrect. What should you inspect first?

1. Production traces and logs
2. The browser theme
3. The application logo
4. The operating-system wallpaper

Correct Answer: A
Tracing helps identify the actual prompt, context, tool calls and model response involved.

Exam Takeaway: Understand the purpose, use case, and trade-off of each concept before memorizing answers.

Final Revision Tip: Scenario-based questions usually test whether you can choose a safe, reliable, maintainable, and cost-effective developer approach.