YugabyteDB MCP

Yugabyte's MCP server gives an AI client a deliberately narrow window into a YugabyteDB cluster: it can understand the schema and read data, and it cannot change anything. The summarize_database tool lists every table in the connected database along with schema information and row counts, which gives an agent the map it needs before writing a query. A second capability executes read only SQL against the cluster, so questions phrased in natural language turn into real queries with real results rather than plausible sounding invention. The read only boundary is the design decision that matters. Because no write path is exposed, connecting an assistant to a production cluster cannot result in an accidental UPDATE without a WHERE clause or a dropped table, which is the usual objection to letting a model near a database. The trade off is that schema changes and data loading stay in existing tooling. Setup is straightforward: clone the repository, install dependencies with the uv package manager, add the server to the client configuration file, and supply a connection string through an environment variable that carries database name, host, port, user and password. It works against YugabyteDB clusters whether they are self managed or running on Yugabyte's managed service. The audience is developers and analysts who want to explore a distributed SQL database conversationally, and teams evaluating a Postgres compatible distributed database who want to ask questions of a test dataset without writing SQL by hand.