A simplified implementation of the Model Context Protocol (MCP) server using Elixir's Phoenix Framework.

MCPheonix is an intelligent, self-healing, distributed AI event system using Model Context Protocol and Elixir's Phoenix Framework. It provides a server that implements the Model Context Protocol, allowing AI models to interact with your application data and functionality through a unified interface.

• Server-Sent Events (SSE) stream for real-time notifications
• JSON-RPC endpoint for client requests
• Simple resource system
• Event publish/subscribe mechanism
• Basic tool invocation
• Flux image generation integration
• Dart task management integration
• Extensible MCP server architecture
• Self-healing distributed architecture via Cloudflare Durable Objects
• Edge computing capabilities through Cloudflare Workers

MCPheonix implements a sophisticated self-healing system through its integration with Cloudflare Durable Objects and Workers, creating a distributed, resilient architecture that can withstand failures and ensure continuous operation.

• Durable Objects: Stateful serverless components running on Cloudflare's global edge network that maintain consistency even across failures.
• Edge Distribution: Critical application state is replicated across Cloudflare's global network, ensuring availability even during regional outages.
• Automatic Recovery: If an instance becomes unavailable, the system automatically recreates it with consistent state from durable storage.
• Real-time Communication: WebSocket support enables instant recovery coordination and state synchronization across the distributed system.
• Event-Driven Architecture: Components react to state changes through a publish/subscribe model, allowing the system to self-heal and adapt to failures.

For detailed information on the implementation, see the Cloudflare Integration documentation.

• Elixir 1.14 or higher
• Erlang 25 or higher
• Phoenix 1.7.0 or higher
• Python 3.9+ (for Flux and Dart integration)
• Node.js 18+ (for Dart MCP server)
• Cloudflare account (for Durable Objects integration)

1. Clone the repository

git clone https://github.com/yourusername/mcpheonix.git
cd mcpheonix

2. Install dependencies

mix deps.get

3. Configure the Cloudflare integration

   • Create a Cloudflare Worker using the template in cloudflare/durable-objects-worker.js
   • Deploy it to your Cloudflare account
   • Set the environment variables:
     • CLOUDFLARE_WORKER_URL: URL of your deployed worker
     • CLOUDFLARE_ACCOUNT_ID: Your Cloudflare account ID
     • CLOUDFLARE_API_TOKEN: API token with Workers and DO permissions

4. Configure the Flux integration (if using image generation)

   • Set up the Flux CLI environment as described in the Flux Integration section

5. Configure the Dart integration (if using task management)

   • Set up the Dart MCP server as described in the Dart Integration section

6. Start the server

mix phx.server

The server will be available at http://localhost:4001.

MCPheonix is designed to work with multiple MCP servers. This system includes a flexible infrastructure for integrating custom MCP servers through:

1. Simple JSON Configuration: Define your server settings in priv/config/mcp_servers.json:

{
  "mcpServers": {
    "your_server_id": {
      "command": "/path/to/executable",
      "args": ["arg1", "arg2"],
      "env": {
        "ENV_VAR1": "value1",
        "ENV_VAR2": "value2"
      },
      "tools": {
        "your_tool": {
          "description": "Description of your tool",
          "parameters": [
            { "name": "param1", "type": "string", "description": "Parameter description" }
          ]
        }
      }
    }
  }
}

2. Automatic Server Management: Servers are automatically loaded and managed during application startup.

For comprehensive implementation details, including the Elixir architecture, server lifecycle management, and protocol handling, see the Adding MCP Servers documentation.

• SSE Stream: GET /mcp/stream

  • Establishes a Server-Sent Events stream for receiving real-time notifications
  • Returns a client ID in the response headers

• JSON-RPC: POST /mcp/rpc

  • Accepts JSON-RPC 2.0 requests
  • Client ID can be provided in the x-mcp-client-id header or will be generated if missing