Ariba's data model is wide enough that finding the right table is half the job, which is precisely what the table-and-column discovery tools automate here. Everything stays read-only and local. The gate is the CData driver: it has to be installed and licensed on the same machine before the server can connect to anything.
A local MCP server around the CData JDBC Driver for SAP Ariba Procurement, which maps the procurement suite onto relational SQL models. The model gets table discovery, column discovery, and a read-only query tool. It speaks `stdio`, so the MCP client must run on the same machine.
- The SAP Ariba Procurement tables the driver exposes, as CSV with a header row — `saparibaprocurement_get_tables`
- The columns of a given table, so SQL is written against real field names — `saparibaprocurement_get_columns`
- A SQL `SELECT` against live procurement data — `saparibaprocurement_run_query`
- A `Tables` property in the connection file that limits the exposed set; leave it blank and everything is exposed
An SAP Ariba Procurement account and a licensed CData JDBC Driver for SAP Ariba Procurement — trial licences work, applied with `java -jar cdata.jdbc.saparibaprocurement.jar --license`. Java to run, Maven to build (`mvn clean install` → `CDataMCP-jar-with-dependencies.jar`). The `.prp` connection file names `DriverClass` as `cdata.jdbc.saparibaprocurement.SAPAribaProcurementDriver`, the driver jar path, a tool `Prefix` such as `saparibaprocurement`, and the JDBC connection string built by the driver's Connection String utility; OAuth sources prompt a browser sign-in there. MIT licensed.
Build from source — clone the repository and build it, then point your client at the binary