Defining prompts

Core Purpose

• Prompts are pre-built, well-tested message templates that MCP servers expose to clients
• They allow server authors to provide specialized, high-quality prompts tailored to the server's domain
• Users can accomplish tasks without prompts, but using server-provided prompts gives better results

The Format Command Example

• Feature: slash command support (e. g. , /format)
• Goal: reformat plain text documents into markdown syntax
• Users could manually ask Claude to do this, but a custom prompt ensures better, consistent results

How Prompts Work in MCP

• Defined using the @prompt decorator in the server code
• Include a name and optional description
• Return a list of messages (user and assistant messages) that can be sent directly to Claude
• Messages can include interpolated values (like document IDs)

Implementation Structure

• Use from mcp. server. fastmcp import prompts and import base
• Function receives arguments (e. g. , doc_id) that get interpolated into the prompt
• Return [base. UserMessage(text=prompt_text)] as a list of messages
• Clients retrieve prompts via the MCP inspector or programmatically

Key Advantage

• Server authors invest time upfront to develop, test, and evaluate prompts
• Clients get reliable, specialized prompts without reinventing them
• Prompts are reusable across any client application connecting to the server

Takeaways

• Prompts are pre-built message templates that MCP servers expose to clients, allowing server authors to provide specialized, high-quality prompts tailored to their domain
• Prompts are defined using the @prompt decorator and return a list of messages that can be sent directly to Claude, with support for interpolated values like document IDs
• Server authors invest upfront effort to develop and test prompts, so clients get reliable, specialized prompts that are reusable across any application connecting to the server