A server owner has two needs: general guidance telling any LLM how to use the server's tools well, and a reusable "triage incident" workflow that users start from a menu. Which mapping fits?
Select an answer to reveal the explanation.
Short Explanation
Server-wide advice for the model goes on the server's welcome mat, the instructions field in discovery. A workflow a person kicks off from a menu is a prompt. Keep those two straight and you won't get caught by the swapped options.
Full Explanation
The server/discover result carries supportedVersions, capabilities, serverInfo in _meta, and an optional instructions field holding natural-language guidance for LLMs on how to use the server effectively. That is the place for server-wide usage advice, because it applies to the server as a whole rather than to one interaction. A workflow that users start deliberately is a prompt: prompts are user-controlled templates the server defines and the user selects, often as a slash command. Putting the guidance in serverInfo fails because serverInfo is the server's self-reported name and version for display, logging and debugging, and clients should not change behaviour because of it. Exposing the workflow as a tool also hands the trigger to the model instead of the user. Swapping the two, guidance in a prompt and workflow in instructions, fails because the model would only see the guidance when a user happened to invoke it, and instructions offers no menu entry to start a workflow. Putting guidance in tool annotations fails by concept: annotations are behaviour hints such as readOnlyHint and destructiveHint, not prose for the model, and a resource is application-controlled context rather than a user-started workflow. Exam caveat: instructions is optional. Operational check: call server/discover, confirm instructions is present, and confirm prompts/list includes the workflow.