glama.json

{
  "$schema": "https://glama.ai/mcp/schemas/server.json",
  "name": "knowledge-rag",
  "display_name": "Knowledge RAG",
  "description": "Local RAG system for Claude Code with hybrid search (inverted-index BM25 + semantic vectors), cross-encoder reranking, snippet mode, min_score filtering, and 12 MCP tools. 20 format parsers. Zero external servers.",
  "author": "Lyon.",
  "repository": "https://github.com/lyonzin/knowledge-rag",
  "homepage": "https://pypi.org/project/knowledge-rag/",
  "license": "MIT",
  "runtime": "python",
  "maintainers": ["lyonzin"],
  "tags": ["rag", "search", "knowledge-base", "embeddings", "reranking", "claude-code", "local-ai", "mcp"],
  "categories": ["Knowledge & Memory", "Search", "Developer Tools"],
  "features": [
    "Hybrid search (inverted-index BM25 + semantic vectors + RRF fusion)",
    "Cross-encoder reranking (ms-marco-MiniLM-L-6-v2)",
    "Snippet mode (truncate to ~500 chars, saves ~72% tokens)",
    "min_score filtering (cut low-relevance noise)",
    "Markdown-aware chunking (splits by section headers)",
    "Query expansion (symmetric synonym groups)",
    "File watcher (auto-reindex on document changes)",
    "SSE/HTTP transport (1 server serves N clients)",
    "Lazy-load ONNX embeddings (zero cold-start cost)",
    "Query cache LRU+TTL (instant repeat queries)",
    "20 file formats (MD, PDF, TXT, PY, C, CPP, H, JS, JSX, TS, TSX, JSON, XML, CSV, DOCX, XLSX, PPTX, IPYNB, MQH, MQ4)",
    "12 MCP tools (search, CRUD, URL ingestion, similarity, evaluation)",
    "Incremental indexing with content-hash deduplication",
    "Optional NVIDIA GPU acceleration (CUDA via ONNX)",
    "Zero external servers (ONNX in-process via FastEmbed)"
  ],
  "install": "pip install knowledge-rag"
}