MCP Infrastructure
Modular server/client platform with dynamic plugins, real-time WebSocket communication, and integrated AI capabilities.
Architecture
C++26 server with WebSocket JSON-RPC 2.0, native desktop client with ImGui, web client with Angular, and a dynamic plugin system with hot-reload.
Plugin Catalog
12+ server plugins and 11+ client plugins covering business operations, development tools, and data processing. Loaded dynamically via dlopen.
AI Hub
Unified AI backend integrating local LLMs via llama.cpp, speech-to-text (Whisper), TTS, image generation (Stable Diffusion), and RAG with vector search.
Business Intelligence
PowerBI-inspired reporting engine with interactive charts, cross-filtering, time grain aggregation, and a web viewer. DuckDB as the analytics backend.
API & Integration
WebSocket JSON-RPC 2.0 protocol, REST gateway with dynamic route mapping, role-based access control, and change data capture notifications.
Deployment & Packaging
Modular .deb packaging, private APT repository, plugin hot-reload without server restart, systemd integration, and VNC remote access.
Company Manager
Model any organization as a team of LLM agents. Define reusable role templates with typed capabilities, hire instances into your orgs, and run them as teams on deterministic workflow graphs, with human gates, one WIP board for every run and a supervision team — all on your MCP infrastructure.
This is the platform we install and operate for client organisations under MCP server deployment. The full rationale, from protocol choice to plugin model to production rollout, is written up in our guide to self-hosted MCP infrastructure. How agent teams run on this platform is shown graph by graph in AI agent teams and workflow graphs. The language choice behind the server is argued in Why We Built Our MCP Server in C++26.