The Model Context Protocol (MCP) is a protocol designed to enable real-time communication between MCP clients, including LLMs, AI agents, and AI apps, and MCP servers, such as MindsDB.

MindsDB is an MCP server that enables your MCP applications to answer questions over large-scale federated data—spanning databases, data warehouses, and SaaS applications.

​
Key Features

  • Unified Data Gateway
    MindsDB acts as a unified data gateway that handles complex data operations, freeing AI apps and agents from the need to manage individual data sources or write complex logic. MindsDB enables “one-step” querying across multiple sources with full auditability.

  • Superior User Experience
    MindsDB delivers a polished, secure, and user-friendly experience on top of MCP integrations. Our turnkey solution includes robust security controls, monitoring, governance, and native integrations to ensure compatibility with non-MCP solutions and applications.

  • Advanced AI Workflows
    MindsDB orchestrates complex AI workflows, such as joining data from multiple sources and coordinating across various models and data systems in a single query. These composite workflows are challenging to implement using MCP alone.

​
Getting Started

MindsDB MCP API runs on port 47337 by default.

To enable the MCP API, you can use the MINDSDB_APIS environment variable:

export MINDSDB_APIS='http,mysql,mongodb,postgres,mcp'

Or, specify it when starting MindsDB:

python -m mindsdb --api http,mysql,mongodb,postgres,mcp

​
Protocol Overview

MCP provides a bidirectional communication channel that allows MCP clients, such as LLMs, AI agents or apps, to run queries over federated data provided by MCP servers, such as MindsDB.

Here, federated data is data that is stored in different databases, applications, and/or cloud providers.

MindsDB MCP API allows you to query federated data as if it were a single database.

​
How It Works

Once you have MindsDB running, you can connect all your data sources and query them as if they were a single database. Then you can plug those databases to your AI applications/clients that use the MCP API.

  1. The client connects to the MCP API.
  2. The client sends a query to the MCP API.
  3. The MCP API sends the query to the federated data source.
  4. The federated data source returns the results to the MCP API.
  5. The MCP API returns the results to the client.

Was this page helpful?

Suggest editsRaise issue