Home › Study guides › CCAR-F › Domain 2

CCAR-F · Domain 2 of 5 · 5 lessons · about 107 min

Domain 2: Tool Design & MCP Integration

Tool descriptions Claude can choose between, errors an agent can act on, the right tools for each agent, and MCP servers wired into Claude Code. 18% of CCAR-F.

A tool is a function Claude is allowed to call: look up an order, search the web, read a file. Claude never sees the code behind a tool. It sees the tool's name, its description, and the schema of the inputs it accepts, and it chooses between tools on that text alone. That single fact explains most of this domain: when an agent picks the wrong tool, the first thing to fix is almost always the description.

MCP, the Model Context Protocol, is the open standard for packaging tools so that Claude Code, the Agent SDK and other clients can discover and call them. An MCP server exposes tools (things to do), resources (things to read) and a standard way of reporting errors. The exam does not ask how to host a server; it asks how to configure one, how to describe its tools, and what a well-formed error looks like.

This domain is 18% of the exam, about 11 questions, drawn mostly from the customer support scenario (tools like get_customer, lookup_order, process_refund), the research system (search and analysis tools split across subagents) and the developer productivity scenario (the built-in Read, Write, Edit, Bash, Grep and Glob tools).

The recurring test: is the fix in the description, the error, the distribution, or the configuration? Four different kinds of answer, and the question tells you which one it wants by describing a specific failure. Learn to read the symptom.

What the exam guide tests

The official CCAR-F guide lists 5 task statements for this domain. Exam questions are written against them, and so are the lessons: each row says where it is covered.

Lessons

1. 2.1 Tool descriptions that route correctlyWhy thin tool descriptions make an agent pick the wrong tool, what a full one contains, when to rename or split a tool, and how prompt wording overrides it.22 min
2. 2.2 Structured error responses for MCP toolsWhy a generic "Operation failed" wrecks an agent's judgement: the MCP isError flag, four error categories, isRetryable, and empty result versus access failure.22 min
3. 2.3 Distributing tools across agents and configuring tool_choiceWhy a long tool list hurts selection, how to scope each subagent to its role, when a cross-role tool fits, and what tool_choice auto, any and forced guarantee.20 min
4. 2.4 Integrating MCP servers: scopes, secrets and resourcesProject .mcp.json versus user ~/.claude.json, ${GITHUB_TOKEN} expansion so no secret is committed, all servers' tools at once, and resources as catalogues.21 min
5. 2.5 Selecting built-in tools: Read, Write, Edit, Bash, Grep, GlobGrep searches contents, Glob matches paths, Edit needs a unique anchor and falls back to Read plus Write, and why an agent explores a codebase incrementally.22 min

Practice question

From the CCAR-F bank, tagged to this domain. Every answer option is explained. Nothing is stored, nothing to sign up for.