Knowledge RAG

What it does

Knowledge RAG is a high-performance, 100% local Retrieval-Augmented Generation (RAG) system that turns your local documents (PDFs, Markdown, code, notebooks) into a searchable knowledge base for AI agents. It enables Claude Code and other MCP clients to natively search your private data without any external API keys or cloud servers.

Tools

• search_knowledge: Hybrid search combining semantic vectors and BM25 keywords with reranking.
• get_document: Retrieve full content of a specific indexed document.
• add_document: Add a new document to the knowledge base from raw content.
• add_from_url: Fetch, clean, and index content from a web URL.
• reindex_documents: Trigger an incremental or full rebuild of the search index.
• get_reindex_status: Poll the progress of background indexing operations.
• list_documents: List all indexed documents, optionally filtered by category.
• list_categories: Show available document categories and counts.
• get_index_stats: Get detailed statistics on chunks, models, and cache hit rates.
• remove_document: Delete a document from the index and optionally from disk.
• update_document: Update the content of an existing indexed document.
• search_similar: Find documents semantically similar to a reference file.
• evaluate_retrieval: Measure retrieval quality using MRR@5 and Recall@5 metrics.

Installation

{
  "mcpServers": {
    "knowledge-rag": {
      "command": "npx -y knowledge-rag"
    }
  }
}

Supported hosts

Confirmed support for Claude Code, Claude Desktop, Cursor, Antigravity, Windsurf, VS Code (Copilot MCP), and Zed.