MCP Prompts

General definition

MCP Prompts are reusable message templates that a Model Context Protocol server publishes so a user can pick one and start a well-formed task. The specification calls them user-controlled: they exist to be explicitly selected, typically as slash commands or menu items in the client, rather than chosen by the model on its own. A prompt has a name, an optional title and description, and an optional list of arguments, each with a name, a description and a required flag.

The protocol methods are:

• prompts/list: discover the available prompts, paginated
• prompts/get with a name and argument values: returns the expanded messages
• each message carries a role (user or assistant) and content that is text, an image, audio or an embedded resource
• notifications/prompts/list_changed when the set changes (servers declare listChanged)

The embedded-resource content type is what makes prompts more than canned text: a prompt can pull a documentation page, a code sample or a configuration file into the conversation at the moment it is invoked, so the assistant starts with the right reference material already in context. Argument values can be auto-completed through the completion API. Servers should validate arguments; a missing required argument is a JSON-RPC invalid params error (-32602).

Contrast with the other two primitives: tools are model-controlled actions, resources are application-controlled context, prompts are user-controlled starting points. A prompt is also different from a system prompt: it is a one-off, user-triggered message, not the standing instructions of an assistant. Not every client surfaces prompts, so a server that relies on them for onboarding usually offers the same material through tools or resources as well.

In the Ethora ecosystem

The Ethora MCP server ships six prompts: ethora-vite-quickstart, ethora-nextjs-quickstart and ethora-backend-sdk-quickstart (wire the chat component or the backend SDK into a fresh project), ethora-auth-map (which token type each route accepts), ethora-recipes (the ordered tool calls for common jobs such as broadcast, source ingestion and the website widget) and ethora-agents-quickstart (create an AI agent, invite it to a room, talk to it). In a client that lists prompts they appear as selectable commands; in one that does not, the ethora-help tool returns the same recipes as a tool result.

They are documented next to the tools on the reference page. The quickstarts point at the Chat SDK packages, so a developer can go from an empty Vite or Next.js project to a working chat screen without leaving the assistant, and the agents quickstart is the fastest route to a live AI agent in a room.