Local RAG Search

What it does

MCP Local RAG provides a fully private, local-first Retrieval-Augmented Generation (RAG) system. It allows AI assistants to ingest and search technical documents, PDFs, and Markdown files without sending data to external APIs.

Tools

• ingest_file: Indexes a local file (PDF, DOCX, TXT, MD) into the vector database.
• ingest_data: Indexes HTML content provided by the assistant.
• query_documents: Performs semantic search with keyword boosting for exact technical terms.
• read_chunk_neighbors: Retrieves surrounding context for a specific search result chunk.
• list_files: Lists all currently indexed files and their status.
• delete_file: Removes a specific file from the index.
• status: Returns system health and database statistics.

Installation

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "local-rag": {
      "command": "npx",
      "args": ["-y", "mcp-local-rag"],
      "env": {
        "BASE_DIR": "/path/to/your/documents"
      }
    }
  }
}

Supported hosts

Confirmed for Claude Desktop, Cursor, Codex, and Claude Code.