This document outlines the architecture and implementation approach for building a scalable, distributed Model Context Protocol (MCP) server using Cloudflare Workers. The system demonstrates how to create a production-ready MCP implementation that can scale globally with minimal operational overhead.
This guide is heavily adapted from the excellent work by @scyto.
Original gist and foundational research: https://gist.github.com/scyto/76e94832927a89d977ea989da157e9dc
The Thunderbolt hardware detection, kernel module loading, systemd configuration, and udev automation techniques are based on scyto's pioneering work on Thunderbolt networking with Proxmox. This guide adapts those proven methods for Proxmox VE 9's new native SDN OpenFabric capabilities.
This builds upon excellent foundational work by @scyto.
- Original TB4 research from @scyto: https://gist.github.com/scyto/76e94832927a89d977ea989da157e9dc
- My Original PVE 9 Writeup: https://gist.github.com/taslabs-net/9f6e06ab32833864678a4acbb6dc9131
Key contributions from @scyto's work:
- TB4 hardware detection and kernel module strategies
A Discord bot that combines AI models (Anthropic Claude, OpenAI GPT, and Meta Llama via Cloudflare Workers AI) with channel-based document search, built entirely on Cloudflare's edge platform. Share knowledge within Discord channels with role-based access control - if you can see the channel, you can access its AI and documents.
Imagine you're at a client meeting. They ask about specific terms in a contract you uploaded months ago. Instead of awkwardly searching through emails or saying "I'll get back to you," you simply pull out your phone, open Discord, and type: