MindsDB
by mindsdb
MindsDB runs an MCP server built into its own query engine, so an agent can list connected sources and run federated SQL across 200+ of them through one query tool.
Run with Docker
docker run --name mindsdb_container -p 47334:47334 mindsdb/mindsdb
About
MindsDB isn't a separate MCP add-on. The server is built into the MindsDB engine itself, reachable once you run the platform (a single docker run starts it) at an endpoint like http://127.0.0.1:47334/api/mcp/v1, with a status check at /mcp/status. Rather than exposing one tool per data source, it exposes a small, general tool set: list_databases to see what's connected, and query to run MySQL-syntax SQL against any of them, because the federation work of talking to Slack, Salesforce, Jira, or a Postgres warehouse already happens inside MindsDB before the query ever reaches the agent.
On top of plain federated SQL, MindsDB's knowledge base tools (search_kb, insert_kb) turn any connected source into something closer to a RAG index: insert_kb ingests data from a source into a knowledge base, and search_kb runs hybrid semantic-plus-keyword search over it. Enterprise setups can put OAuth 2.1 in front of the MCP endpoint instead of relying on the default open-access mode.
Key features
- MCP server built directly into the MindsDB engine, one install, no separate MCP package
- query tool executes MySQL-syntax SQL against any of 200+ connected federated sources
- list_databases enumerates every data source currently connected to the instance
- Knowledge base tools (insert_kb, search_kb) add hybrid semantic + keyword search over ingested data
- Optional OAuth 2.1 in front of the MCP endpoint for enterprise deployments
- Single Docker container to run the whole engine, including the MCP endpoint
Use cases
- Letting an agent answer questions that require joining data across several disconnected business systems
- Turning scattered docs, tickets, or messages (Slack, Jira, Gmail) into a searchable knowledge base for RAG
- Giving an agent safe, read-oriented SQL access to a data warehouse without a hand-built custom connector
- Standing up one federated query layer instead of maintaining a separate MCP server per data source
Available tools
list_databases
Lists every data source currently connected to the MindsDB instance.
query
Executes a MySQL-syntax SQL query against one or more connected data sources.
insert_kb
Ingests data from a connected source into a knowledge base for semantic search.
search_kb
Runs hybrid semantic and keyword search over a knowledge base.
Frequently asked questions
Do I need a separate MCP package to use MindsDB with an agent?
No. The MCP server is part of the core MindsDB engine. Running the engine (via Docker or otherwise) exposes the MCP endpoint alongside the normal query interface.
Can it query more than SQL databases?
Yes. MindsDB's federation layer covers SaaS apps and unstructured sources too (Slack, Gmail, Jira, Salesforce among them), all reachable through the same query and knowledge-base tools.