MCP Prompt
What is an MCP prompt?
An MCP prompt is a named prompt template exposed by an [MCP server](/glossary/mcp-server/). An [MCP client](/glossary/mcp-client/) can discover available templates, request one by name, and supply defined arguments to customize the returned messages.
MCP prompts help servers package reusable interaction patterns for the domains they support. They are typically user-controlled: an application may present them as selectable commands or workflow starters.
How it works
A server declares support for prompts and publishes prompt metadata, such as a name, description, and optional arguments. When a client requests a prompt, the server returns structured prompt messages built from the supplied values.
For example, a repository server might provide a review_changes prompt with arguments for branch name and review focus. The returned messages could guide a model to inspect selected resources and produce a review.
The server provides the template; the host decides how to display it, whether to combine it with other context, and when to send the resulting messages to a model.
MCP prompt versus a normal LLM prompt
A [prompt](/glossary/prompt/) is any input that guides a model. An MCP prompt is a protocol-exposed, discoverable prompt template with an MCP-defined retrieval flow. Once retrieved, its content still becomes ordinary model-facing messages within the host’s larger prompt construction.
An MCP prompt is also not automatically a system prompt. It can return structured messages with supported roles, but the host controls final message assembly and model invocation.
MCP prompt versus agent instructions
[Agent instructions](/glossary/agent-instructions/) define persistent behavioral guidance for an agent. An MCP prompt usually represents an optional, reusable interaction pattern selected for a particular task.
A client should not silently treat server-provided prompt text as higher-priority trusted instructions. Prompt content comes from an external integration and should be handled according to the host’s trust and security model.
Why it matters
MCP prompts let domain integrations share useful workflows without hard-coding every template into each AI application. They can improve discoverability and consistency while keeping the user in control of the selected task.
Their quality depends on clear naming, purposeful arguments, safe content handling, and avoiding hidden assumptions about the host’s model or system instructions.