Add MCP Prompts support

[7tg]

Description

The MCP protocol supports Prompts as a primitive for providing templated messages/instructions to AI agents. This package currently only implements Tools.

MCP Prompts Overview

Prompts are reusable templates that help agents understand how to interact with the system. They can include:

• Pre-defined instructions for common operations
• Contextual guidance based on model metadata
• Best practices for CRUD operations

Proposed Prompts

Discovery & Understanding

• explore_models - Guide agent through discovering available models and their relationships
• understand_model - Detailed walkthrough of a specific model's fields, constraints, and admin config

CRUD Operations

• create_guide - Step-by-step guidance for creating objects with validation hints
• update_guide - Best practices for updating objects, handling inlines
• bulk_operations_guide - How to efficiently perform bulk create/update/delete

Relationships

• navigate_relationships - How to traverse and understand model relationships
• manage_inlines - Working with inline objects effectively

Admin Actions

• execute_action - Guide for discovering and executing admin actions safely

Implementation

1. Add prompts to ServerCapabilities
2. Implement prompts/list handler
3. Implement prompts/get handler with dynamic argument substitution
4. Create prompt templates (possibly as Django templates or Python strings)

MCP Protocol Reference

# prompts/list response
{
    "prompts": [
        {
            "name": "explore_models",
            "description": "Guide for discovering Django admin models",
            "arguments": []
        },
        {
            "name": "understand_model", 
            "description": "Deep dive into a specific model",
            "arguments": [{"name": "model_name", "required": true}]
        }
    ]
}

# prompts/get request
{"method": "prompts/get", "params": {"name": "understand_model", "arguments": {"model_name": "user"}}}

Benefits

• Reduces agent trial-and-error
• Provides consistent guidance across different AI models
• Self-documenting API usage patterns
• Can be dynamically generated based on actual model configuration

Labels

enhancement, feature