MCP Startup Boilerplate v0.0.1

A robust, free, ready-to-use Rails boilerplate for building AI-powered applications that integrate seamlessly with Claude AI using the Model Context Protocol (MCP). This boilerplate includes user authentication and OAuth2 provider capabilities out of the box.

🚀 Features

User Authentication: Complete authentication system powered by Devise
OAuth2 Provider: Full OAuth2 server implementation with Doorkeeper
MCP Integration: Seamless Claude AI integration using fast-mcp
Payment Processing: Integrated Stripe payments for tools with subscription options
Modern Rails: Built on Rails 8.0.2 with the latest features
Tailwind CSS: Beautiful, responsive UI with Tailwind
Hotwire: Modern, fast frontend with Turbo and Stimulus

📋 Requirements

Ruby 3.2+
PostgreSQL
Node.js & Yarn (for frontend assets)
Stripe Account (for payment processing)

🔧 Setup

1. Clone the repository

git clone https://github.com/f/mcp-startup-boilerplate mcp-startup
cd mcp-startup

2. Install dependencies

bundle install
yarn install

3. Setup the database

rails db:create db:migrate db:seed

4. Configure Stripe (optional for payment features)

Set your Stripe API keys in your environment:

export STRIPE_PUBLISHABLE_KEY="pk_test_your_key"
export STRIPE_SECRET_KEY="sk_test_your_key"
export STRIPE_WEBHOOK_SECRET="whsec_your_webhook_key"

For development, you can also update these in config/initializers/stripe.rb

5. Start the server

rails server

Your application should now be running at http://localhost:3000

💳 Payment Integration

This application comes with built-in Stripe payment processing for MCP tools:

Subscription Plans

Standard Plan: $0.03 per tool call with 40% savings compared to pay-as-you-go
Enterprise Plan: $0.03 per tool call with additional benefits and unlimited usage

Pay-Per-Use Pricing

Default pricing: $0.05 per tool call (can be customized per tool)

Payment Management

User-friendly subscription management interface
Detailed usage tracking dashboard
Tool usage history and analytics

🧰 Paid Tool Framework

The application includes a PaidTool base class that allows you to easily create tools that require payment:

class YourPaidTool < PaidTool
  description "Your paid tool description"
  
  # Custom pricing (optional, defaults to $0.05)
  def price_cents
    8 # $0.08 per call
  end
  
  def call
    return { error: "User has no active subscription" } unless charge_user
  
    # Your implementation here after charge
  end
end

How PaidTool Works

When a tool is called, it first checks if the user has an active subscription
If subscribed, the tool executes and logs usage without additional charges
If not subscribed, the user is charged the tool's price via Stripe
All tool usage is tracked for analytics and billing purposes

Extending PaidTool

You can customize pricing and behavior by:

Overriding price_cents to set custom pricing for specific tools
Creating different tool categories with varying price points
Adding tool-specific features for premium subscription tiers

🔒 OAuth2 Integration

This boilerplate comes with a complete OAuth2 server implementation powered by Doorkeeper. Key endpoints:

OAuth Authorization: /oauth/authorize
Token Endpoint: /oauth/token
Application Registration: /oauth/applications
Token Revocation: /oauth/revoke

The OAuth server supports:

Authorization Code Flow
Client Credentials Flow
Refresh Token Flow
PKCE Extension (S256 and plain)

🤖 MCP Integration

This boilerplate is pre-configured to work with Claude AI through the Model Context Protocol, enabling AI applications to securely access your application's data and functionality.

Available MCP Resources

users/me - Access information about the authenticated user

Available MCP Tools

MeTool - Get information about the currently authenticated user (do not delete it)
FinancialCalculatorTool - Example paid calculator tool

🔌 Integrating with Claude

To connect your application with Claude:

Register your application in Claude Desktop or other Claude integrations
Configure your MCP server in Claude's configuration:

{
  "mcpServers": {
    "your-server-name": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://your-server-url/mcp/sse"
      ]
    }
  }
}

For Claude Desktop, this file is located at:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

🔧 Customization

Adding New MCP Tools

Create a new tool in app/tools/:

class YourTool < ApplicationTool
  description 'Your tool description'

  arguments do
    required(:param_name).filled(:string).description('Parameter description')
  end

  def call(param_name:)
    # Your implementation
    
    # Access the current authenticated user
    user = current_user
    # Do something with the user data
    { user_id: user.id, result: "Processed data for #{user.email}" }
  end
end

Adding New MCP Resources

Create a new resource in app/resources/:

class YourResource < ApplicationResource
  uri 'your/resource/path'
  resource_name 'Your Resource'
  description 'Your resource description'
  mime_type 'application/json'

  def content
    # Access the current authenticated user
    user = current_user
    
    JSON.generate({
      user_id: user.id,
      email: user.email,
      # Additional user-specific data
      custom_data: user.custom_data
    })
  end
end

Working with Current User

Both MCP Tools and Resources have access to the current_user method, which returns the authenticated user from the OAuth token. This allows you to:

Retrieve user-specific data
Perform user-authorized operations
Maintain security by scoping operations to the authenticated user
Build personalized AI experiences based on user context

For security reasons, always verify user permissions before exposing sensitive data or performing critical operations.

🛡️ Security Considerations

All MCP endpoints require proper OAuth authentication
User data is only accessible to authorized applications
OAuth tokens can be revoked by users at any time
PKCE extension is supported for public clients
Stripe payments are securely processed with modern best practices

f/mcp-startup-boilerplate