MCP prompts let servers offer ready-made prompt templates that users pick, often as commands in the host application.
How They Work
A server lists prompts with names, descriptions and optional arguments. When a user selects one and supplies arguments, the server returns messages that the host adds to the conversation.
Good Uses
- Standard workflows: "review this pull request", "draft a release note", "triage this incident".
- Domain expertise: templates encoding how your organisation does something.
- Combining context: a prompt can include resource content, such as a ticket's details, automatically.
Why Not Just Tools?
Prompts are user-initiated: the user chooses to start a workflow. Tools are model-initiated. Prompts suit things people explicitly want to kick off.
Design Tips
- Name prompts after the task.
- Keep arguments few and clearly described.
- Write templates as carefully as any production prompt, and test them.
Security
Prompt content goes straight into the conversation. Hosts should make clear what a prompt inserts, and users should trust the server providing it.