MCP Prompts

Prompts are template-based messages that language models can use. Each prompt has a name, description, arguments, and message templates.

Prompt Structure

Every prompt in ModelContextProtocol.jl is represented by the MCPPrompt struct:

• name: Unique identifier for the prompt
• description: Human-readable explanation of the prompt's purpose
• arguments: List of parameters the prompt accepts
• messages: Template messages with placeholders for arguments

Creating Prompts

Here's how to create a basic prompt:

Note: The role field uses the Role enum with values user and assistant. You may need to import these:

import ModelContextProtocol: user, assistant

greeting_prompt = MCPPrompt(
    name = "greeting",
    description = "Personalized greeting message",
    arguments = [
        PromptArgument(
            name = "name",
            description = "User's name",
            required = true
        ),
        PromptArgument(
            name = "time_of_day",
            description = "Morning, afternoon, or evening",
            required = false
        )
    ],
    messages = [
        PromptMessage(
            role = user,
            content = TextContent(
                text = "Hello! {?time_of_day?Good {time_of_day}}! My name is {name}."
            )
        )
    ]
)

Arguments

Prompt arguments are defined using the PromptArgument struct:

• name: Parameter identifier
• description: Explanation of the parameter
• required: Whether the argument must be provided (default: false)

Template Syntax

Prompt templates support parameter substitution and conditional blocks:

• Basic substitution: {parameter_name}
• Conditional blocks: {?parameter_name?content if parameter exists}

Registering Prompts

Prompts can be registered with a server in two ways:

1. During server creation:

server = mcp_server(
    name = "my-server",
    prompts = my_prompt  # Single prompt or vector of prompts
)

1. After server creation:

register!(server, my_prompt)

Directory-Based Organization

Prompts can be organized in directory structures and auto-registered:

my_server/
└── prompts/
    ├── greeting.jl
    └── faq.jl

Each file should export one or more MCPPrompt instances:

# greeting.jl
using ModelContextProtocol

greeting_prompt = MCPPrompt(
    name = "greeting",
    description = "Personalized greeting message",
    arguments = [
        PromptArgument(name = "name", description = "User's name", required = true)
    ],
    messages = [
        PromptMessage(
            role = user,
            content = TextContent(text = "Hello! My name is {name}.")
        )
    ]
)

Then auto-register from the directory:

server = mcp_server(
    name = "my-server",
    auto_register_dir = "my_server"
)

Advanced Examples

Multi-Message Conversation Prompt

conversation_prompt = MCPPrompt(
    name = "code_review",
    description = "Code review conversation template",
    arguments = [
        PromptArgument(
            name = "language",
            description = "Programming language",
            required = true
        ),
        PromptArgument(
            name = "code",
            description = "Code to review",
            required = true
        ),
        PromptArgument(
            name = "focus_area",
            description = "Specific area to focus on",
            required = false
        )
    ],
    messages = [
        PromptMessage(
            role = user,
            content = TextContent(
                text = "Please review this {language} code:{?focus_area? Focus on {focus_area}.}"
            )
        ),
        PromptMessage(
            role = user,
            content = TextContent(text = "```{language}\n{code}\n```")
        ),
        PromptMessage(
            role = assistant,
            content = TextContent(
                text = "I'll analyze this {language} code{?focus_area? with focus on {focus_area}}."
            )
        )
    ]
)

Prompt with Image Content

visual_prompt = MCPPrompt(
    name = "analyze_diagram",
    description = "Analyze a diagram or chart",
    arguments = [
        PromptArgument(name = "image_path", description = "Path to image", required = true),
        PromptArgument(name = "question", description = "Question about the image", required = false)
    ],
    messages = [
        PromptMessage(
            role = user,
            content = TextContent(
                text = "Please analyze this diagram{?question?: {question}}"
            )
        ),
        # Note: In practice, you'd load the actual image data
        PromptMessage(
            role = user,
            content = ImageContent(
                data = UInt8[],  # Placeholder for image bytes
                mime_type = "image/png"
            )
        )
    ]
)