Exam Scenarios/Practice Scenarios
Advanced15 min

Scenario: Developer Productivity with Claude

Optimize a development team's workflow with Claude Code and the Agent SDK. Covers codebase exploration strategies, MCP server scoping, tool distribution for agents, session management, and scratchpad patterns.

Quick Reference

  • →Glob for finding files by name, Grep for searching content, Read for known file paths
  • →Project-scoped MCP servers (.mcp.json) vs user-scoped (~/.claude.json) serve different purposes
  • →Limit agents to 4-5 tools for focused tasks — more tools increase selection errors
  • →Session resumption preserves context but may carry stale assumptions after code changes
  • →Scratchpad files let agents persist findings across tool calls without consuming context window

Scenario Description

You are building a suite of developer productivity tools using the Claude Agent SDK and Claude Code. The target environment is a mid-size engineering organization with 15 active repositories, ranging from a large legacy Java monolith to modern TypeScript microservices. Developers need help with four categories of tasks: codebase exploration (understanding unfamiliar code), boilerplate generation (scaffolding new services and components), automated code review (pre-merge checks), and legacy system documentation (generating docs for undocumented code).

Your tools connect to MCP servers for repository access, CI/CD integration, and a custom knowledge base of internal architecture decision records (ADRs). Key design decisions include: which built-in tools to expose for each task category, how to scope MCP servers for team vs. individual use, how to keep agent tool sets focused, and when to start fresh sessions vs. resuming previous ones. The team includes developers of varying experience levels, so the tools must be intuitive enough for juniors while powerful enough for seniors.