What is MCP prompt?

Prompts are the MCP feature for packaging instructions. A server that supports them declares the prompts capability. A client calls prompts/list to see what is available, then prompts/get with a prompt name and any arguments. The server responds with a list of messages, each with a role of user or assistant. Message content can be text, an image, audio, a resource link or an embedded resource, so a prompt can pull server-managed documents into the conversation.

Each prompt definition has a name and may include a title, a description, icons and a list of arguments, some of which can be marked required. Argument values can be auto-completed through the completion API.

The specification calls prompts user-controlled. That describes who decides when a prompt is used, not who wrote its content, which is still defined by the server. The typical presentation is a user-initiated command, with slash commands given as the example. This contrasts with tools, which the model chooses to call, and resources, which the application decides how to include.

Errors follow JSON-RPC conventions: an invalid prompt name or a missing required argument returns an invalid params error. The spec's security section requires implementations to validate what goes into and comes out of prompts, to guard against injection and unauthorized resource access.