MCP Prompts are reusable sets of instructions or context that you can provide to Large Language Models (LLMs) when they interact with your MCP Servers. They help guide the LLM’s responses, define its persona, or set specific rules for a task.

Creating Prompts

  1. Open the Create Modal: Click the “Create” button in the app sidebar or press ‘C’.
  2. Select “Prompt”: In the modal, choose the “Prompt” option.
  3. Name Your Prompt: Give the prompt a descriptive name (e.g., “Customer Service Persona”, “Code Explainer Instructions”).
  4. Write the Prompt Content: Enter the actual text of the instructions or context you want to provide to the LLM.
  5. Save the Prompt: Click “Create” or “Save”.

Your prompt will now appear in your library of prompts.

Associating Prompts with Servers

To make a prompt available to one or more MCP Servers:

  1. Go to the Prompts Table: Navigate to the main “Prompts” section from the sidebar.
  2. Find Your Prompt: Locate the prompt you want to associate in the table.
  3. Open Edit Servers Menu: Click the three-bar menu icon on the prompt’s table row and select ‘Edit Servers’.
  4. Select Servers: In the modal that appears, check the boxes next to the servers you want this prompt to be available for.
  5. Save Associations: Click “Save”.

Now, the selected servers (and the LLMs using them) can potentially access this prompt.

Using Prompts

How prompts are used depends on the client application interacting with the MCP Server:

  • MCP Developer Chat: Currently, the Developer Chat application does not support selecting or applying MCP Prompts during conversations (coming soon).

  • Local MCP Clients (e.g., Claude Desktop): Client applications like Claude Desktop can leverage prompts. When interacting with an MCP Server that has prompts associated with it, the client application may allow you to select one or more of those prompts to be included in the context sent to the LLM for the current conversation or task.

Prompts provide a powerful way to standardize LLM instructions and behavior across different interactions connected to your MCP Servers.