Obsidian + Claude = no more copy paste

<!-- SC_OFF --><div class="md"><p>I gave Claude persistent memory across every session by connecting Claude.ai and Claude Code through a custom MCP server on my private VPS. Here’s the open source code.</p> <p>I got tired of Claude forgetting everything between sessions. So I built a knowledge base server that sits on my VPS, ingests my Obsidian vault, and connects to Claude Code and Claude.ai through MCP.</p> <p>The result: when I write something in Claude.ai, Claude Code can instantly search and read it. When Claude Code captures a terminal session with bugs and fixes, I can access that knowledge from Claude.ai in the next conversation. Same brain, different interfaces.</p> <p>But it goes further. I also built a multi-agent orchestrator called Daniel that wraps Claude, Codex, and Gemini CLIs. All three share the same knowledge base. When Claude hits rate limits or goes down (like it did yesterday), Codex picks up with the same context. Zero downtime.</p> <p>The self-learning part: every session, the AI automatically updates its own instruction files based on what worked and what didn’t. After 100+ sessions, the AI knows my codebase, my preferences, my architecture patterns. It one-shots clean code because it’s accumulated enough context.</p> <p>Google just open-sourced their Always-On Memory Agent two weeks ago. Mine’s been running in production with multi-agent orchestration and human curation that theirs doesn’t have.</p> <p>Both projects are open source:</p> <pre><code>∙ Knowledge Base Server (the brain): https://github.com/willynikes2/knowledge-base-server ∙ Agent Orchestrator (Daniel): https://github.com/willynikes2/agent-orchestrator </code></pre> <p>Tech stack: Node.js, SQLite FTS5, MCP, Express, Obsidian Sync. No vector database, no cloud dependencies. ~$60/month for three premium AI agents with persistent memory.</p>