Developer Foundations · 10.6% of the exam

tools and mcps

This domain is about a tenth of the scored questions. It covers how to define, describe and approve the tools Claude calls, how to build a server for the Model Context Protocol and choose its transport, and how to pick the right home for a new capability.

Course example

This domain keeps following Shelfwise, the internal stock-question tool, which is a long-lived Python service on the Agent SDK. Its agent answers shop staff from one stock feed through stock_mcp, its one MCP server: a program that offers tools and resources to Claude applications over MCP, the Model Context Protocol.

In Domain 8 Shelfwise gains its tools and the server behind them: one read-only tool, stock_lookup, allowed by name, and a reservation tool designed on paper and parked behind staff approval (8.1). stock_mcp is written with the Python SDK and launched as a subprocess inside the service's container, with a stock://feed-status resource that engineers attach in Claude Code (8.2). The service uses no built-in tools and no Skills, because the stock rules that every request needs stay in the system prompt (8.3).

stock_lookup: read-only, allowed by namestock_reserve: parked, staff approvestock_mcp: a subprocess, no porttools=[], setting_sources=[]
1 toolin Claude's context, stock_lookup, inside the cached ≈6,000-token prefix. The wildcard mcp__stock__* in the first allow list would have approved every reservation with no prompt. A draft feed_status tool would have joined that prefix and let Claude choose it in a staff chat.

Start hereThe words used throughout this domain

Messages API
The Claude API endpoint that takes a conversation and returns Claude's next turn, keeping no state between requests (2.3).
Agent SDK
The Anthropic library that runs Claude in a tool loop inside your own program (1.2).
tool_use, tool_result
Claude's block that asks for a tool call, and your code's block that returns the outcome (1.2).
hook, PreToolUse
Your code that runs at a fixed point in the agent loop. A PreToolUse hook runs before a tool call and can block it (7.3).
is_error, isError
The flag that marks a failed tool call's result: is_error on a Claude API tool_result block, isError on an MCP tool result (8.1).
JSON-RPC
The message format MCP follows: each request names a method, such as tools/call, and carries an id that its response repeats (8.2).
URI
Uniform Resource Identifier: a string that names a resource, such as a web address or stock://feed-status (8.2).
OAuth 2.1
The authorization framework MCP specifies for HTTP servers that require authorization: the client gets an access token for the user from an authorization server and sends it with every request (8.2).
Skill
A folder of instructions built around SKILL.md, whose short description lets Claude match a task to it and whose body loads only when used (2.5, 8.3).
plugin
A directory that Claude Code installs and enables as one unit, holding Skills, hooks, subagents and MCP servers (2.5). 8.3 covers writing one.

task statements

flashcards

0 / 9 known
Card 1 of 9