Prompts (MCP)

Prompts are one of the three primitives an MCP server can expose, alongside tools and resources. A prompt is a named, parameterized template the server defines — "summarize-pr", "analyze-table", "triage-error" — that a client can fetch, fill in with arguments, and hand to the model as a starting message.

Crucially, prompts are user-controlled by design: the person picks one to invoke, typically from a menu or as a slash command, rather than the model deciding on its own.

A client calls prompts/list to discover what a server offers — each entry has a name, a description, and a declared set of arguments. When the user invokes one, the client calls prompts/get with the argument values, and the server returns fully rendered messages ready for the model's context.

Because the server renders the template, it can do real work at fetch time: pull in current data, embed a resource, or tailor the message to the arguments — the "template" can be as dynamic as the author wants.

Most MCP servers have a right way to be used — a sequence of tool calls, a framing that yields good results — and prompts let the author encode that knowledge into the protocol itself instead of hoping users read the README or the model infers it.

A database server, for example, might ship an "explain-schema" prompt that instructs the model to inspect tables before writing queries. Without it, every user rediscovers that workflow by trial and error; with it, the best practice is one slash command away.

The three MCP primitives differ in who initiates them. Tools are model-controlled: the LLM decides to call them mid-task. Resources are application-controlled: the client chooses what data to load into context. Prompts are user-controlled: a human explicitly picks a template to start or steer a workflow.

A useful shorthand — tools are verbs the model can use, resources are nouns the client can read, and prompts are recipes the user can invoke.

Prompts themselves carry no payment semantics — in x402-monetized servers it's the tool calls that are priced and paid per call. But prompts shape the economics indirectly: a well-designed prompt steers agents toward efficient tool sequences, which matters to whoever is paying per call. When evaluating a server on Loomal's index, shipped prompts are a good signal that the author has thought about how their tools should actually be used.