Three tools — list tables, list columns, run a SELECT — is the smallest surface that lets a model answer real analytical questions, and the query tool executing SELECT only is what makes pointing it at a production warehouse defensible. CData names the alternatives for anyone who needs more: their downloadable MCP server for full write access, and Connect AI for a managed remote endpoint. The build cost is real, and the Tables property is where you decide how much of the warehouse is in scope.
A local, read-only MCP server from CData that wraps their JDBC driver for Amazon Redshift. The driver exposes Redshift as relational SQL models and the server puts three tools over it, so a model can find the tables, check the columns and run a query without SQL being written by hand.
- `redshift_get_tables` — the tables available in the data source, returned as CSV with a header row
- `redshift_get_columns` — the columns on a named table, so field names are checked rather than guessed
- `redshift_run_query` — executes a SQL SELECT query
- Scope control through the `Tables` property: blank for everything, or an explicit list of what to expose
The CData JDBC Driver for Amazon Redshift, installed and licensed separately — `java -jar cdata.jdbc.redshift.jar --license` takes your name, email and TRIAL or a licence key. Java to run and Maven to build: `mvn clean install` produces `CDataMCP-jar-with-dependencies.jar`. Then a `.prp` file with `Prefix=redshift`, a server name and version, the driver path, `DriverClass=cdata.jdbc.redshift.RedshiftDriver`, the `JdbcUrl` from the driver's connection-string utility, and `Tables`. The client runs java with the jar and the .prp path. Transport is stdio, so client and server share a machine. MIT licensed.