For the past few months, I kept hitting the same wall with Claude: every conversation started from zero. If I wanted it to reference something I'd read or a decision I'd made weeks ago, I had to copy-paste it in myself or worse, it would just guess and get it wrong.
So I built Neurix MCP, a small server that connects my personal knowledge graph (notes, research, past decisions) directly to Claude. When I ask something related to my own context, Claude calls a tool to search it semantic search across English and Vietnamese, no copy-pasting. Writes go through a preview confirm step, so nothing gets saved without me approving it first.
It's solo-built, I use it daily, and it's genuinely changed how I work with Claude answers feel grounded in what I actually know instead of what the model assumes.
Curious how others here are thinking about this problem: are you managing personal/AI context manually, or have you tried something similar (RAG setups, other MCP servers, etc.)? What's worked, what hasn't?