Ingests and manages contextual information and training data through multiple iterations, performing multi-pass lexical retrieval over a chunk graph without a vector database.
MicroSearch functions as a versatile Model Context Protocol (MCP) server designed for efficient knowledge retrieval. It allows users to ingest contextual information and training-style JSONL data, maintaining multiple named iterations on disk. Its core retrieval mechanism employs a unique four-tool ladder—preview, retrieve, and MCP handoff—operating over a BDH-style lexical chunk graph that supports 'next'/'prev' link expansion, all without requiring a vector database. This makes it an ideal backend for AI agents and other systems needing robust, iteration-aware information access.
