Engineering blog
Articles from the I-Machine team on self-hosted MCP infrastructure, agentic AI in the enterprise, and the architecture of our plugin platform.
Self-Hosted MCP Infrastructure for Enterprise : Architecture, Plugins, and Deployment
Why and how to run your own MCP infrastructure : Protocol choices, plugin architecture, local LLMs, agentic loops, and shipping to production.
Why We Built Our MCP Server in C++26
Most MCP servers are written in TypeScript or Python. Here is why ours runs on C++26 with GCC 16, and what it bought us.
MCP vs REST : Protocol Design for AI Agents
Why MCP uses WebSocket JSON-RPC 2.0 instead of REST, and when that choice actually matters for production AI agents.
Hexagonal Architecture for MCP Plugins
How the I-Machine plugin platform applies ports-and-adapters across 40+ plugins and two front-ends from a single application layer.
Plugin Hot-Reload in C++ : Three dlopen Constraints
The lock discipline, asynchrony contract, and reload verification that hot-reloadable C++ plugins require in production.
C++26 Reflection in Production : Using P2996 With GCC 16
P2996 is mergeable today on GCC 16. Here is what we run in production with it.