Back to Home
RajwardhanMali icon

MDBQS

by RajwardhanMali

Overview

A multi-database query system enabling natural language querying of heterogeneous databases (SQL, NoSQL, Graph, Vector) via LLM-assisted planning and parallel execution with result fusion and provenance tracking.

Installation

Run Command
./run-all-servers.sh

Environment Variables

  • POSTGRES_DSN
  • MONGO_URI
  • MONGO_DB
  • NEO4J_URI
  • NEO4J_USER
  • NEO4J_PASSWORD
  • MILVUS_HOST
  • MILVUS_PORT
  • GEMINI_API_KEY
  • APP_ENV

Security Notes

The system utilizes an LLM (Google Gemini) to directly generate database query strings (SQL, NoSQL filters, Cypher components) which are then executed without robust semantic validation or sanitization. This presents a high risk of prompt injection and subsequent database injection (SQL injection in PostgreSQL, potential NoSQL injection for MongoDB via '$where'/'$eval', Milvus query injection in `get_metadata` via f-string for `cust_id`). Although the SQL adapter claims 'ONLY SELECT allowed', this is an LLM instruction and not technically enforced, making it vulnerable to malicious query chains. Hardcoded database credentials (e.g., 'postgrespassword', 'neo4jpassword', 'minioadmin') are present in `docker-compose.yml` for development, which is a severe risk if used in production.

Similar Servers

Stats

Interest Score0
Security Score3
Cost ClassMedium
Avg Tokens1000
Stars0
Forks0
Last Update2025-12-06

Tags

Multi-DatabaseLLMFastAPIQuery EngineData FederationProvenance Tracking