MCP Agent OS.
Production-ready MCP architecture reference implementation
Overview
A production-ready MCP architecture reference implementation demonstrating how to build AI agents that consume and expose Model Context Protocol servers. This project demonstrates the MCP chaining pattern.
Features
| Feature | Description |
|---|---|
| MCP Consumer | Connects to GitHub, Phoenix Docs, and Search MCP servers |
| MCP Provider | Exposes Agent OS as an MCP endpoint |
| Arize Tracing | Full observability with OpenTelemetry integration |
| Multi-Team Support | Specialized clients for PM, DevRel, Sales, and Engineering |
| Session Memory | SQLite-backed conversation history and summaries |
How It Works
1
MCP Consumer
Connects to external MCP servers - GitHub, Phoenix Docs, and Search - using HTTP and command-based transports.
2
Agent OS
Orchestrates specialized agents with session memory (SQLite-backed) and full observability via Arize OpenTelemetry tracing.
3
MCP Provider
Exposes the Agent OS as an MCP endpoint at http://localhost:7777/mcp for downstream clients.
4
Multi-Team Clients
Specialized clients for PM, DevRel, Sales, and Engineering teams - each querying the same Agent OS with different perspectives.
Tech Stack
PythonAgno FrameworkFastAPISQLiteAnthropic APIArize OpenTelemetry