Defining prompts

Overview

• Prompts are the final major feature in MCP servers (after resources and tools)
• They allow server authors to expose pre-built, well-tested prompts to client applications
• Users can already accomplish tasks manually, but custom prompts provide better, more reliable results

Purpose of Prompts

• Define high-quality, specialized prompts ahead of time that are tailored to the server's domain
• Prompts have been evaluated and tested to work across various scenarios
• Expose these prompts so clients don't need to develop their own or guess at optimal wording
• Example: a document management server can provide a "format document" prompt instead of leaving users to write their own instructions

How Prompts Work

• Prompts are defined using the @prompt decorator (similar to tools and resources)
• Each prompt has a name and optional description
• Prompts accept arguments (like a document ID) that can be interpolated into the prompt text
• When a client requests a prompt, the server returns a list of messages (user and assistant messages)
• These messages can be sent directly to Claude for processing

Implementation Structure

• Use @prompt decorator with name and description
• Function receives arguments (e. g. , doc_id: string)
• Return a list of messages containing the pre-written prompt
• Arguments are interpolated into the prompt text before returning

Example: Format Document Prompt

• Accepts a document ID as input
• Instructs Claude to read the document using available tools
• Asks Claude to rewrite content in markdown syntax
• Optionally instructs Claude to save changes back to the document

Takeaways

• Prompts are pre-built, tested, domain-specific instructions that MCP servers expose to clients, eliminating the need for users to write their own prompts
• Prompts are defined using the @prompt decorator and accept arguments that get interpolated into the prompt text
• When a client requests a prompt, the server returns a list of messages that can be sent directly to Claude for processing
• Prompts provide better, more reliable results than manual task completion by leveraging server authors' expertise and testing