Applications and Integration
Free Claude Certified Developer (CCDV-F) practice — 6 questions on Applications and Integration, with explanations. No sign-up. Full 12-question mixed test →
A developer is building a document analysis pipeline that processes 50,000 PDF files overnight using the Messages API. Each document requires vision analysis and generates approximately 8,000 tokens of output. The pipeline runs once per week. To minimize cost while meeting the overnight processing window, which approach is MOST cost-effective?
Message Batches API offers 50% cost reduction compared to standard API calls and is explicitly designed for high-volume, non-realtime workloads like overnight processing. Prompt caching further reduces costs by reusing system prompts across the 50,000 requests. The overnight window (no realtime requirement) makes batching ideal.
A Claude Desktop application is behaving differently than expected when compared to the same prompts tested on claude.ai. The developer notices that tool calls are being made automatically in Desktop but require explicit user confirmation on claude.ai. What is the MOST likely cause of this difference?
Claude Desktop supports MCP servers which can provide tools to Claude with pre-authorized access based on server configuration. This is a key architectural difference: Desktop has local integration capabilities through MCP that claude.ai (browser-based) does not have. MCP servers define which tools are available and their permission model.
A developer implements prompt caching for a RAG application that provides customer support documentation to Claude. The system prompt contains 45,000 tokens of documentation that changes once per day. The application receives approximately 1,000 user queries per day. After implementing caching with a 5-minute TTL, costs are higher than expected. What is the PRIMARY issue?
With 1,000 queries per day, queries arrive roughly every 90 seconds on average. A 5-minute TTL means the cache frequently expires between queries, requiring cache writes (which cost the same as regular tokens) without achieving sufficient cache hits. For this usage pattern, the default TTL (5 minutes minimum, up to hours) should be extended, or documentation should be structured to maximize reuse. Cache writes cost full price; only cache reads provide savings.
An application using Claude's vision capabilities needs to analyze medical imaging scans. The developer is choosing between base64-encoded image data in the Messages API versus using the Files API. Each scan is 8MB, analyzed once, and must be processed within 30 seconds. Compliance requirements mandate that image data never persists on external servers beyond the analysis request. Which approach BEST meets these requirements?
Base64-encoded images in Messages API requests are processed ephemerally — they are not stored server-side after the request completes. This directly satisfies the compliance requirement that images never persist on external servers. The 30-second processing requirement is achievable with direct API calls (Files API doesn't provide faster processing for vision tasks).
A development team is implementing retry logic for Claude API calls. During testing, they observe that some requests fail with HTTP 529 (overloaded) errors during peak hours. The application currently implements exponential backoff starting at 1 second with a maximum of 5 retries. Users report that the application feels unresponsive during these retry cycles. What is the BEST improvement to the retry strategy?
HTTP 529 indicates the API is overloaded. Continuing to retry aggressively worsens the overload. A circuit breaker pattern detects repeated failures and fails fast, preventing long retry cycles that make the application feel unresponsive. Displaying a clear error message improves UX. 3 retries with exponential backoff is reasonable before opening the circuit. This balances resilience with user experience and API health.
A developer is building a code review assistant that uses Claude Code via the API. The assistant needs to understand the full repository context including file structure, dependencies, and coding patterns. The repository contains 150 files totaling 2.5MB of code. Which approach provides the MOST comprehensive context to Claude while optimizing for cost and performance?
CLAUDE.md files are specifically designed to provide project context to Claude-powered tools like Claude Code. The file is automatically read and provides high-level architecture, patterns, and navigation guidance. Combined with dynamically including relevant files per request (based on what's being reviewed), this provides comprehensive context without sending unnecessary code. This is the recommended pattern for repository-aware Claude applications.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$54.99 $17.99 with code FREETEST33 — valid through September 1.
Get my $17.99 deal →