> ## Documentation Index
> Fetch the complete documentation index at: https://docs.0mcp.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Prompts

> Create reusable prompt templates for MCP clients.

A **prompt** is a reusable message template that helps AI clients start common tasks with consistent instructions.

Unlike [**tools**](/capabilities/tools), prompts don't perform actions. Instead, they provide structured instructions that guide the AI model. Prompts can also accept user inputs, allowing the same template to be reused with different values.

<Note>
  A prompt guides the AI model but does not call your API by itself. Use a tool when the workflow must perform an action.
</Note>

## When to use prompts

Prompts are useful for repeatable AI workflows, such as:

* Summarizing support cases
* Drafting customer emails
* Writing release notes
* Analyzing logs
* Reviewing pull requests
* Generating product descriptions

Instead of rewriting the same instructions every time, clients can invoke a prompt and provide the required inputs.

## Creating a prompt

Each prompt includes:

* Name
* Title
* Description
* Arguments, if needed
* One or more messages

Messages can have either the **user** or **assistant** role and may contain plain text or reference existing [resources](/capabilities/resources).

## Arguments

Arguments make prompts reusable.

Define the values a client must provide when invoking the prompt, such as a case ID, customer name, or product identifier.

Argument names:

* May contain letters, numbers, and underscores (`_`)
* Must begin with a letter or underscore
* Can be marked as required or optional

Reference an argument using Go template syntax:

```text theme={null}
Summarize the support case for {{.case_id}}.
```

When the prompt runs, `{{.case_id}}` is replaced with the value provided by the client.

## Best practices

* Give prompts clear, task-oriented names.
* Keep each prompt focused on a single workflow.
* Use arguments instead of hardcoded values whenever possible.
* Write instructions that produce consistent results across different inputs.

## Publishing changes

Prompts become available to MCP clients only after you [publish a new version](/guides/versioning) of your MCP server.

Draft prompts remain private until they're published, allowing you to review and test changes before making them available.