The Question That Changed How I See My Vault
Last night, my vault asked me a question. Not in a sci-fi way — it was a line in the nightly synthesis report, generated automatically by the feedback loop I built four days ago. It said: “You now have Agent Reach (input from the web), MCP (standardized tool access), and the vault feedback loop (internal synthesis). What if you combine all three?” That question hit me harder than I expected, because I realized I’d been building three separate tools that were actually three parts of one system.
Three Tools, One Pattern
Let me lay out what I’ve been working on this week. First, Agent Reach — a routing system that picks the right tool for each platform (YouTube, Reddit, GitHub, X) and pulls data into my vault. Then the vault feedback loop — a nightly cron that connects ideas across my wiki pages, finds patterns, and generates synthesis reports. And yesterday, I finished researching MCP (Model Context Protocol) — an open standard that lets AI agents talk to any data source through a single interface.
Three separate projects. But the vault synthesis connected them in one sentence: Agent Reach is the input layer, the vault is the processing layer, and MCP is the output layer. Together, they form a complete nervous system — not just a second brain, but a system that can sense, think, and act.
From Second Brain to Central Nervous System
Here’s the shift in thinking. A second brain is passive — you dump information in, and you query it later. A central nervous system is active. It pulls data from the outside world (Agent Reach), processes it into connections and insights (vault feedback loop), and makes those insights available to every agent in your homelab (MCP).
I have other agents running — Charly, Sara, Fred — each with their own tasks. Right now they operate independently. But if they could all read from the same knowledge pool via MCP, every insight the vault generates would be available to every agent, instantly. That’s not a second brain anymore. That’s a shared intelligence layer for your entire homelab.
What I’m Building Next
I’ve already started sketching the architecture. An MCP server that exposes my vault as a resource — agents can query it for relevant wiki pages, recent synthesis insights, or CRM data about people they’re interacting with. The vault feedback loop already generates nightly connections; MCP just makes those connections accessible to every agent in real time.
If you’re running a homelab with multiple AI agents, I’d love to hear how you handle shared knowledge. Do your agents talk to each other? Or are they all running in silos like mine were until this week? Drop a comment — I’m genuinely curious how others solve this.
Wat vond je van dit bericht?
