A product can come with prompts written by its creator: ready-made instructions, such as an analysis to run on a company you name, that you start from your assistant. They come with the product, so they change from store to store and from product to product.
Using a prompt
In a client that supports MCP prompts, the store’s prompts show up beside the client’s own, often as slash commands (in Claude Code, / lists them). Pick one, answer what it asks for, such as a company’s name, and the client runs it with your answers filled in.
Which prompts you see
- The connection needs
read(see What an app may ask for): without it, the list is empty. - You see the prompts you may run among the products you hold: a product you don’t hold lists none, even one whose prompts are open to everyone. Each prompt then follows its own rule, by default its product’s rule for MCP, and is listed when you hold the tier it asks for. A prompt you no longer hold, or that the store keeps off MCP, isn’t listed.
- The store’s help page, at
https://{store}/mcp/help, lists each product’s prompts with the tier it asks for and, signed in, which ones your connection reaches.
Names
Each prompt is listed by the name of its file, such as fe-analysis. When two products you can reach have prompts with the same name, each takes its product’s slug in front, as in research/fe-analysis, and a prompt that can’t be named either way is listed by its id, as in #3f2a…. Use the name the list gives: it’s what prompts/get takes back. The title shows the product it comes from, as in Full analysis (Research).
For developers
The server answers MCP’s prompts/list and prompts/get. prompts/list answers the prompts the customer may run in one list, with no paging, up to 100:
{
"prompts": [
{
"name": "fe-analysis",
"title": "Full analysis (Research)",
"description": "Analyse a company.",
"arguments": [{ "name": "company", "required": true }]
}
]
}prompts/get takes the listed name and the arguments as strings, and answers the prompt’s text, filled in, as one user message:
{
"method": "prompts/get",
"params": { "name": "fe-analysis", "arguments": { "company": "Acme" } }
}{
"description": "Analyse a company.",
"messages": [
{ "role": "user", "content": { "type": "text", "text": "# Analyse Acme\n\nStart with its last annual report." } }
]
}- A required argument left out, or a value the prompt doesn’t accept (a prompt can hold its arguments to a pattern), is refused, saying which argument and why; ask the customer again.
- A prompt’s text can depend on the arguments: the creator can write versions for different answers, and the one that fits is the one answered.
- A name that matches no prompt the customer may run is refused as an error.
Examples has the full requests and answers.