The questions that come up about a service desk — how many tickets of this kind are open, which accounts keep coming back — are usually a report someone else builds. Exposed as SQL tables they become a conversation. This build has no write path: three tools, all read. If you need to create or update records, CData's separate MCP server product covers that; here the absence is the safety property.
A local, read-only MCP server from CData for Oracle Service Cloud. The CData JDBC driver presents the service data as relational tables; this server wraps the driver so an assistant can discover tables, read columns, and run SELECT queries against live records — the natural-language question goes in, the SQL is composed for you.
- The tables available in Oracle Service Cloud, as CSV with a header row — `get_tables`
- The columns on a table, so a query is written against the real schema — `get_columns`
- A SQL SELECT run against live data — `run_query`
- A prefix on the tool names taken from your `.prp` file, so multiple CData servers coexist in one client config
- A `Tables` setting that restricts which tables are reachable, instead of exposing the whole source
A licensed CData JDBC Driver for Oracle Service Cloud, installed and licensed separately, plus Java. Build from a clone with `mvn clean install` to get `CDataMCP-jar-with-dependencies.jar`. The `.prp` file carries the prefix, server name and version, driver path, driver class, the JDBC connection string produced by the driver's Connection String utility, and the optional table list; sources using OAuth authenticate in a browser during that step. Transport is stdio — the server runs on the same machine as the client.
Build from source — clone the repository and build it, then point your client at the binary