MCP Project Template

Spin up a production-ready MCP server via cli helper, optionally ingest your project planning doc.

What's This?

This repo contains everything you need to prompt a local LLM (like Claude Code) to generate a complete MCP server from scratch. Think of it as a smart template that your AI assistant can read and implement.

Default project layout:

project-root/
├── src/
│   ├── core/
│   │   ├── server.js          # Enhanced MCP server with middleware
│   │   ├── client.js          # MCP client with retry logic
│   │   ├── middleware.js      # Request/response interceptors
│   │   └── base-tool.js       # Base tool class
│   ├── tools/
│   │   ├── registry.js        # Tool registry with versioning
│   │   ├── index.js           # Tool exports
│   │   └── [domain]/          # Domain-specific tools
│   │       └── [tool-name].js
│   ├── services/              # Business logic
│   ├── config/
│   │   ├── index.js           # Configuration management
│   │   └── schemas.js         # Config validation
│   ├── errors/
│   │   └── index.js           # Error taxonomy
│   ├── observability/
│   │   ├── logger.js          # Structured logging
│   │   └── metrics.js         # Metrics collection
│   └── utils/
│       └── validation.js      # Shared validators
├── test/
│   ├── fixtures/
│   └── test-utils.js          # Testing utilities
├── examples/                  # Example implementations
├── package.json
├── tsconfig.json             # Optional TypeScript config
└── .env.example

The Pieces

• MCP-PROJECT-TEMPLATE.md - The full architectural blueprint with code examples
  • Code entities intentionally live in this doc rather than being pre-made at the root so you can configure via consulting with your LLM assistant on parts you do and don't want to generate.
• MCP-SEED-PROMPT-TEMPLATE.md - A fill-in-the-blank prompt template
• MCP-SEED-PROMPT-EXAMPLES.md - Real examples of prompts that work
• setup.js - Interactive CLI to generate custom prompt.

How to Use

0. (Optional) Planning doc setup - Fill out planning.md with your MCP project details
1. Run the setup CLI - npm run setup
2. Answer the prompts - Tell it what domain you want to build and any specific requirements
3. Get your custom prompt - The CLI generates a tailored prompt and copies it to your clipboard
4. Feed it to your LLM - Paste into Claude Code, Cursor, or whatever local AI you're using
5. Get a working MCP server - Your AI will generate a complete implementation following the production-ready patterns

Why This Is Helpful

Instead of starting from scratch, your LLM gets:

• A complete architectural pattern with middleware, error handling, and observability
• Working code examples for tools, services, and configuration
• Testing utilities and best practices
• Clear patterns for scaling and maintaining the codebase

Quick Start

Option 1: Use the CLI (recommended)

npm run setup

Option 2: Manual setup

# Copy the seed prompt template
cp .setup/MCP-SEED-PROMPT-TEMPLATE.md my-prompt.md

# Edit it with your domain info
# Then paste the whole thing into your AI assistant

This will allow your LLM assistant to generate a complete MCP server with proper structure, error handling, and all the production-ready stuff you'd want but don't want to write by hand.

Examples

Check out MCP-SEED-PROMPT-EXAMPLES.md for specific prompts that generate:

• GitHub repo management servers
• Data processing pipelines
• File system automation
• API monitoring tools
• Multi-service integrations

Fire up your local LLM assistant in this repo, paste in an example, and 💸 watch those tokens fly 💸

TODO

• Minimize LLM calls
  • Determine minimal common source code entities, prepopulate stubs
  • More granualar config in setup tool
  • "hackathon" -> ... -> prod levels of generation (observability stuff, etc prob not needed for personal projects)
    • Identify features to build in
• Copy initial generated prompt to local file as well as system clipboard
• TypeScript template
• Other language templates
  • Extend setup script to nuke node-specific parts for non-JS projects