Knowledge

Michael Kanda's Knowledge Base server connects Claude and other LLMs directly to designare.at's RAG pipeline, the same infrastructure that powers his AI assistant Evita. It exposes two tools: search_knowledge for semantic search across his entire knowledge base using Gemini embeddings and Upstash Vector, and get_services which returns structured JSON of his web and AI services without any HTML parsing. The server runs as a Vercel serverless function using streamable HTTP transport and responds in a median 426ms. It's a practical example of GEO on the protocol level, where instead of waiting for crawlers to parse your site, you give LLMs a direct line to your knowledge base with structured, controlled data.