Frictionless capture across devices. Structured tools that let agents actually work with my knowledge.
A self-hosted Model Context Protocol (MCP) system that gives agents structured, executable access to my personal knowledge — continuously fed by low-friction capture across all my devices.
The real power isn't just the vault. It's the combination of seamless input (voice, notes, whatever) flowing into a structured knowledge base, plus 47+ typed skills that let agents do real work instead of guessing from context.
It is production infrastructure. I use it every day across four synced devices.
The foundation is a structured, long-term knowledge base built from years of work, research, decisions and systems. What makes it powerful is how little friction there is to keep it alive.
Capture happens naturally (voice, quick notes, whatever). Agents don't get vague RAG — they get typed tools with explicit schemas, security boundaries, and real actions they can take.
This is the piece that makes the vault usable by agents. Built with Starlette, it exposes real capabilities through the Model Context Protocol.
I wanted agents that actually know my context and can act on it — not generic RAG over scattered documents. The combination of frictionless capture + structured knowledge + typed MCP skills makes the whole thing usable in real work.
This is also real production experience running reliable agent infrastructure with the Model Context Protocol, outside of any big tech platform.
Python + Starlette
OAuth 2.0 PKCE
47+ typed skills
Systemd + Nginx
Multi-device sync
I'm building this in public because the space is new and most people are still figuring out what good MCP infrastructure actually looks like in production.
A minimal starter template is coming (Starlette + OAuth PKCE + 3 example skills).