MCP Prompts
Overview
Apollo MCP Server supports MCP prompts — reusable prompt templates that AI clients can discover and use. Prompts are guided workflows that help AI models use your GraphQL tools more effectively.
No configuration is required. The server automatically detects a prompts/ directory in the working directory and advertises prompt support to MCP clients.
Configuration
Configure the prompts directory in your config file:
prompts:
path: /config/promptsThe path can be absolute or relative to the process working directory. Defaults to prompts. Set via APOLLO_MCP_PROMPTS__PATH environment variable.
Defining Prompts
Store .md files in a prompts/ directory:
your-project/
├── config.yaml
└── prompts/
└── check_orders.mdPrompt File Format
Each file has YAML frontmatter and a template body separated by ---:
---
name: check_orders
description: "Look up a user's recent orders and summarize them"
arguments:
- name: email
description: "The customer's email address"
required: true
- name: limit
description: "Maximum number of orders to return"
---
1. Use GetUserByEmail(email={{email}}) to find the user
2. Use GetUserOrders to retrieve their recent orders{{limit}}
3. Summarize each order with its status, total, and dateFrontmatter Fields
| Field | Type | Required | Description |
|---|---|---|---|
name |
string |
Yes | Unique identifier for the prompt |
description |
string |
No | Human-readable description for AI clients |
arguments |
list | No | Arguments the prompt accepts |
Argument Fields
| Field | Type | Default | Description |
|---|---|---|---|
name |
string |
— | Argument name, used as {{name}} placeholder |
description |
string |
— | Description shown to AI clients |
required |
boolean | false |
Whether the argument must be provided |
Template Placeholders
Use {{argument_name}} in the template body. The server replaces each placeholder with the value provided in a prompts/get request. Unmatched placeholders remain unchanged.
Behavior
- If the
prompts/directory does not exist, the server starts normally without prompt support - Required arguments are validated; missing ones return an error
- Each prompt returns a single user-role text message
- Multi-message and non-text content types are not currently supported