GrooveSeek
An MCP server for hybrid semantic and full-text search over a Markdown knowledge base, with optional support for plain text, PDF, Office documents and source code.
GrooveSeek parses Markdown files with YAML frontmatter, and optionally .txt, .pdf, .docx, .xlsx, .pptx files plus Rust, Python or PHP source, into heading-based chunks (or one chunk per definition for source code). It generates embeddings with FastEmbed by default (BGE-small-en-v1.5, or BGE-M3 for multilingual and Japanese content) and stores them in SQLite with sqlite-vec for vector search alongside FTS5 for full-text search, with optional cross-encoder reranking.
A live-sync file watcher keeps the index current across manual edits, git pulls and external scripts, and an optional TOML schema can validate frontmatter conventions and let searches filter on custom fields.
It connects to Claude Code, Cursor or any MCP-compatible client over stdio or Streamable HTTP transport. Releases before 1.0.0 are beta with no compatibility guarantee.