Master the vocabulary of the Model Context Protocol — tools, resources, prompts, and the roles of hosts, clients, and servers.
0 / 5 completed
1 / 5
During a standup, a teammate says the AI assistant can't access your internal wiki because the MCP server wasn't granted the right resource. What does an MCP resource represent?
In MCP, resources are structured data or content (files, DB rows, wiki pages) exposed by the server. Tools are callable functions, and prompts are reusable templates.
2 / 5
In a PR review, a comment questions whether the new MCP tool should be defined on the host or the server. Which role actually defines and exposes tools?
The MCP server registers and exposes tools. The host launches servers and passes their capabilities to the model; the client is the protocol session inside the host.
3 / 5
Your team is wiring up a prompt in MCP so users can invoke a structured workflow. A colleague asks what differentiates an MCP prompt from a tool. The correct answer is:
MCP prompts are reusable, parameterised message templates that users select explicitly. Tools are functions the model calls autonomously during inference.
4 / 5
A bug report says the IDE plugin loses context after reconnect because the MCP client drops the session. In MCP architecture, what is the client responsible for?
The MCP client manages one protocol session (JSON-RPC over stdio or SSE) with a single server. The host can run multiple clients, one per server.
5 / 5
During architecture review, you propose using MCP sampling so the server can request the model to generate text. A skeptic asks when a server would need sampling. The best answer is:
MCP sampling lets servers ask the host to run a model generation mid-flow — enabling server-side agentic steps where the server drives LLM calls rather than just responding.