@florentine-ai/mcp

What it does

This server translates natural language questions into database queries for MongoDB and MySQL, returning structured results directly to your AI agent or application. It performs automatic schema exploration so agents understand what data is available, supports semantic vector search and retrieval-augmented generation (RAG) workflows for embeddings-based lookups, enforces tenant-level data isolation for multi-tenant deployments, and lets you exclude sensitive fields from queries. The flow is straightforward: your agent poses a question, the server transforms it into a database query, executes it, and returns the results for further processing.

Who it's for

Backend engineers building custom AI agents that need conversational access to MongoDB or MySQL. Development teams integrating Claude or other LLMs into applications backed by relational or document databases. Organizations building multi-tenant systems where strict data isolation between customers is required.

Common use cases

• Query MongoDB or MySQL from an AI agent using natural language instead of SQL
• Implement semantic search over database records using vector embeddings and RAG
• Build multi-tenant applications with automatic data isolation per tenant
• Answer analytical questions about your database without writing custom queries

Setup pitfalls

• Requires a Florentine.ai account (free tier available) and an active API key—the npm package alone will not work.
• Bring-your-own-key model: you must supply an LLM API key (OpenAI, Google, Anthropic, or Deepseek) via the LLM_KEY environment variable or save it in your Florentine.ai dashboard.
• Your MongoDB or MySQL database must already be connected to Florentine.ai and at least one collection or table must be analyzed and activated in the dashboard before queries will function.
• Requires Node.js v18.0.0 or later; earlier versions are not supported.