Defining prompts

MCP Server Prompts: Key Concepts

**I. Purpose of Prompts in MCP Servers**

• Goal: To allow users to leverage highly specialized, pre-tested, and evaluated prompts tailored to the server's specific capabilities.
• Value Proposition: While a user can manually prompt the system (e. g. , "Convert this to markdown"), a custom, server-defined prompt ensures the best possible result for a specific task (e. g. , document reformatting).
• Function: Prompts allow the server authors to define a set of high-quality, reliable instructions that clients (like a CLI) can use without needing to develop the prompt themselves.

**II. Implementation via Slash Commands**

• Feature: Adding support for slash commands (e. g. , /format) within the MCP server.
• Workflow:
• User types /.
• The application lists supported commands (e. g. , format).
• User selects a command (e. g. , format) and is prompted for necessary arguments (e. g. , document_ID).
• The server executes the defined prompt using the provided arguments.

**III. Defining a Prompt in the MCP Server**

• Mechanism: Prompts are defined using a specific syntax similar to tools and resources.
• Structure:
• Use the prompt decorator.
• Assign a name and an optional description to the prompt.
• Implement the function to receive arguments (e. g. , doc_ID).
• The function must return a list of messages (user and assistant messages) that will be sent to the cloud model.
• Execution Flow (Example: Reformatting):
• The prompt instructs the model (Claude) to take a document_ID.
• The model implicitly uses a tool (e. g. , read_document) to fetch the content.
• The model rewrites the content using the specified format (e. g. , Markdown).
• The model uses another tool (e. g. , edit_document) to save the updated content.

**IV. Key Takeaways**

• Specialization: Prompts allow the server to be specialized (e. g. , document management, editing).
• Abstraction: Prompts abstract complex, high-quality instructions away from the end-user and the client application.
• Testing: Prompts are designed to be well-tested and evaluated before being exposed to clients.

Takeaways

• MCP server prompts allow users to leverage specialized, pre-tested instructions for specific tasks, ensuring high-quality results without manual prompting.
• Prompts are implemented through slash commands (e. g. , /format), where the server executes a defined prompt using arguments provided by the user.
• Defining a prompt involves using a decorator and implementing a function that returns a list of messages for the cloud model to process.
• Prompts abstract complex logic and tool usage (like reading or editing documents) away from the end-user and the client application.