dljones555/MinimalMcpDemo

minimalmcpdemo 🚀

A minimal end-to-end demo for the Model Context Protocol (MCP) using .NET and the official .NET MCP SDK. This project demonstrates how to serve, discover, and consume LLM prompts over the MCP protocol, with advanced prompt management via PromptLoader.

🏗️ Project Structure

• McpPromptServer:
  A minimal MCP server that exposes prompt files (*.prompt.md and more) via HTTP/SSE using the MCP protocol.

  • PromptLoader integration: Advanced prompt management, folder-based organization (roots), prompt chaining, versioning, and filtering.
  • Resources & root support: Prompts are grouped by root (top-level folder) and exposed as MCP resources for discoverability and filtering.
  • Lists available prompts and resources.
  • Serves prompt content on request using a fluent API.

• McpPromptClient:
  A console client that connects to the MCP server, lists available prompts/resources, fetches prompt content, and interacts with an LLM endpoint (e.g., Azure OpenAI) using those prompts.

  • Supports root/resource selection and prompt filtering.

• PromptLoader:
  A reusable library for loading, organizing, and composing prompts.

  • Fluent API: Easily load prompts from folders/files, combine/chains prompts, and apply configuration in a readable style:

            var ctx = await PromptContext
                .FromFolder()
                .WithConfig(config)
                .LoadAsync();

            string combined = ctx.Get("Sales").CombineWithRoot().AsString();

• Configurable via appsettings.json
• Supports prompt versioning (use Git versioned files), chaining, and multi-format (Markdown, YAML, Jinja, etc).

📦 Key Technologies

• .NET 9
• ModelContextProtocol (.NET MCP SDK)

📝 Usage

1. Start the MCP Prompt Server

dotnet run --project McpPromptServer

Prompts are served from the Prompts/ directory (or as configured in appsettings.json).

2. Set your LLM API key

Set the GH_TOKEN environment variable to your LLM provider API key (e.g., Azure OpenAI):

• Windows (cmd): set GH_TOKEN=your_api_key
• Windows (PowerShell): $env:GH_TOKEN="your_api_key"
• Linux / macOS: export GH_TOKEN=your_api_key

3. Run the Client

dotnet run --project McpPromptClient

• Use /resources to see available prompt roots/resources.
• Use /root [name] to select a root.
• Use /listprompts to see available prompts in the current root.
• Use /getprompt [name] to fetch a prompt.
• Type your message to chat with the LLM using the selected prompt.
• Type /exit to quit.

📂 Prompt Format & Configuration

Prompts can be Markdown (*.prompt.md), plain text, YAML, Jinja, etc. Place them in the Prompts/ or PromptSets/ directory, or configure folders in appsettings.json.

Example appsettings.json PromptLoader section:

"PromptLoader": {
  "PromptListType": "named",
  "PromptList": ["system", "instructions", "examples", "groundings", "guardrails", "output"],
  "ConstrainPromptList": true,
  "PromptsFolder": "Prompts",
  "PromptSetFolder": "PromptSets",
  "SupportedPromptExtensions": [".txt", ".prompt", ".yml", ".jinja", ".jinja2", ".prompt.md", ".md"],
  "PromptSeparator": "\n{filename}:\n---\n",
  "CascadeOverride": true

• PromptListType: How to interpret the prompt list: named, numeric, none.
• PromptList: Specific named prompts to load/combine
• ConstrainPromptList: Boolean. Only load prompts in the list
• PromptSetFolder/PromptsFolder: Where to find prompt sets/files
• SupportedPromptExtensions: File types to load
• PromptSeparator: How to join prompts when combining
• CascadeOverride: Whether deeper folders override root prompts by same name (e.g., system prompts)

🤝 References

• Model Context Protocol Spec
• .NET MCP SDK on NuGet

✨ Why?

This project is a minimal, hackable reference for anyone looking to:

• Build or integrate MCP-compliant prompt servers/clients.
• Experiment with advanced prompt management, roots/resources, and LLM orchestration in .NET.
• Use or extend a fluent API for prompt ops.
• Learn about the MCP and prompt ops ecosystem.