Kairox Arch Prep
EXAM SCENARIO

Developer productivity with Claude

What the developer productivity scenario tests on the CCA-F exam: MCP server design, transports, tool consolidation, scoping and shared config.

The developer productivity scenario is about connecting Claude to the systems a development team already uses — repositories, issue trackers, documentation, databases — so engineers can explore and operate them faster. It sits at the meeting point of tool design and MCP integration with Claude Code configuration.

The setup you should picture

A team works in Claude Code. Some capabilities run locally next to each developer, such as access to local files or a local build. Others are shared services, such as company documentation or an analytics database, served by central MCP servers. Project configuration makes the same servers available to everyone working in the repository.

Decisions the exam probes

Transport choice. A local server that a client launches as a subprocess uses stdio and exposes nothing on the network. A shared remote server uses the HTTP-based transport and needs authentication and per-user authorization.

Tool shape. Dozens of narrow tools that mirror database tables waste context and confuse selection. Fewer task-oriented tools with clear names, descriptions, parameter formats and enums work better. Overlapping tools should be merged with an explicit mode parameter.

Output size. A tool that returns thousands of rows floods the context window. Filters, pagination, limits and counts keep results useful.

Least privilege. If Claude should only read analytics data, the server connects with a read-only credential. Tool descriptions that say "read-only" do not make anything read-only.

Sharing configuration. Servers the whole project needs go in the project-scoped MCP configuration committed to the repository, with credentials supplied through environment variables or each user's own sign-in rather than written into the file.

Common traps

How to practise

Build a tiny MCP server with two overlapping tools and watch Claude choose between them. Rewrite the descriptions until the choice is reliable, then merge the tools and compare. Add a list tool that returns a large result set and introduce pagination. Finally, register the server in a project configuration file and confirm a teammate's setup would pick it up without any secret in the file.

The six scenarios are listed in Anthropic's official exam guide (version 1.1), with four drawn per sitting. The emphasis described here is our own study guidance, not official exam content.

Practice questions set in this scenario

Original questions, not real exam items. All 15 questions in this scenario also appear in the practice exam.

Practice question 1

Domain 1: Agentic Architecture & Orchestration

The task: classify each incoming invoice, extract five fields, and store them. The steps never change. Should you build a fixed workflow (predefined steps in code that call Claude) or an autonomous agent (Claude decides which tools to call and when)?

  1. A. A multi-agent system with one specialised agent responsible for each extracted field
  2. B. An autonomous agent, because it can adapt automatically if the invoice process changes
  3. C. A fixed workflow, because known, stable steps are cheaper and easier to test that way
  4. D. An autonomous agent that writes a step-by-step plan for each invoice before it acts
Show answer and explanation

Answer: C. Use the simplest pattern that solves the problem. When the sequence of steps is known and stable, a deterministic workflow gives lower cost and latency and is easier to test. Agent autonomy earns its keep when the path cannot be predicted. Per-invoice planning or one agent per field adds overhead without benefit here.

Practice question 2

Domain 1: Agentic Architecture & Orchestration

Which situation most clearly justifies splitting work across multiple specialised subagents instead of using a single agent?

  1. A. The system prompt has grown long, and splitting it across agents would shorten each copy
  2. B. The single agent sometimes sounds unsure, and several agents could vote on the answer
  3. C. The work spans several large, independent areas whose detail would overflow one context
  4. D. The task is short and needs just one tool, but you want each step to be reviewed
Show answer and explanation

Answer: C. Multi-agent designs cost more tokens and add coordination complexity, so they pay off when work is broad, parallelisable and context-heavy: each subagent digs deep in its own context and returns a summary. Short single-tool tasks are simpler with one agent. Prompt length and tone are not architectural reasons to split.

Other scenarios