Prompts

Create a prompt

In the server's Prompts workspace, create a prompt with a name, optional description, and template text. Names start with a letter and may contain letters, numbers, underscores, and hyphens; names are limited to 64 characters. Descriptions can be up to 300 characters and templates up to 50,000 characters.

Use the template to give the client a focused instruction or context. Keep it clear about which values are supplied by the user and which are fixed in the template.

Define arguments

Add named arguments for the values a client should supply. Mark each argument required or optional, and set a default value where appropriate. Argument names use letters, numbers, and underscores and must not be Python keywords.

Use double braces to reference an argument in the template:

Summarize the following report for {{audience}}:
{{report}}

Argument placeholders are substituted as plain text; they are not evaluated as code. MCP clients can discover the prompt and request it with the values the template needs.

Reference workspace values

A prompt may also reference a workspace variable or secret with ${NAME}. Configure values under Settings → Variables & secrets. Secret values are resolved when the prompt is requested, so the resulting prompt can expose them to the requesting MCP client.

Use prompts in a client

Connect the client to the server and open its prompt picker. The prompt's name and description help the user choose the right template; the client asks for required arguments and sends the filled prompt to the model.

Keep prompt arguments distinct from workspace values: {{argument}} is client-supplied input, while ${NAME} comes from workspace configuration.