PostgreSQL MCP Server

PostgreSQL MCP Server is a Model Context Protocol server that connects an AI assistant to a PostgreSQL database. It was published as one of the reference implementations alongside the protocol specification, and it demonstrates the pattern most database servers still follow: expose the schema as browsable resources, expose querying as a tool, and hold the connection read-only so an agent cannot mutate production data. On connecting, the server enumerates the tables in the database and offers each one as a resource whose contents describe the column names and data types. An assistant reads those resources to learn the shape of the schema before writing anything, which is what prevents the usual failure mode of a model inventing column names that do not exist. The query tool then accepts SQL and executes it inside a read-only transaction, returning rows as structured results. Aggregations, joins, and exploratory filtering all work; inserts, updates, and schema changes are refused by the transaction itself rather than by a prompt instruction that a model might talk itself out of. It suits developers who want to interrogate a database in plain language during development, analysts prototyping queries with an assistant that can see the real schema, and anyone learning how an MCP server is structured. The reference copy sits in the archived Model Context Protocol servers repository, and community and vendor-maintained Postgres servers extend the same idea with richer schema introspection and explicit write modes behind opt-in flags.