course contents
Developer Foundations · 33.1% of the exam
applications and integration
share
This is the heaviest domain on the exam, about a third of the scored questions. It covers the software around each Claude call: the requirements that shape it, the API mechanics of the call itself, the engineering and design of the application that makes it, and how its configuration is kept under control.
This domain follows Shelfwise, the course's own example rather than an exam scenario: an internal tool where staff at a small chain of hardware shops type a stock question into a chat widget and an agent answers from the shops' stock feed. Three engineers run the agent as a long-lived Python service on the Agent SDK, the Anthropic library that runs an agent's tool loop. It reaches the feed only through one small MCP server: a program that offers the feed to the agent as a tool over MCP, the Model Context Protocol.
In Domain 2 Shelfwise gains the Claude API calls behind the widget and the software around them: requirements sized against the platform's limits (2.1), recorded test questions and a staged rollout (2.2), a cached request prefix (2.3), concurrent test runs and a merge gate (2.4), one session per staff chat (2.5), and a pinned model ID and prompt version (2.6).
Start hereThe words used throughout this domain
- token
- The unit Claude reads and writes text in, and the unit the API counts and prices: input tokens for what you send, output tokens for what Claude writes.
- context window
- The most text, counted in tokens, that one request can hold, including the conversation you resend and the room left for Claude's reply.
- Messages API
- The Claude API endpoint that takes a conversation and returns Claude's next turn. Every SDK, including the Agent SDK, calls it underneath.
- system prompt
- Instructions sent in the request's
systemfield, separate from the conversation, that apply to the whole exchange. - tool_use, tool_result
- The block in which Claude asks for a tool call, and the block your code sends back with the outcome.
- prompt caching
- Reusing the unchanged start of a request, its prefix, on later requests at a small fraction of the input price. 2.1 relies on it; 2.3 shows how it works.