Ad reporting is exactly the kind of question where a confident guess is worthless, and the read-only SELECT path here means every figure comes back from the platform rather than from the model's priors. The three-tool surface keeps it simple; the CData driver licence is the prerequisite that decides whether this is practical for you.
A local MCP server built on the CData JDBC Driver for LinkedIn Ads, which presents the ad platform as relational SQL tables. The model is given the table list, the column list, and a read-only query tool, so campaign questions are answered from live data. It runs over `stdio` and therefore only works with a client on the same machine.
- The LinkedIn Ads tables the driver exposes, as CSV with a header row — `linkedinads_get_tables`
- The columns of a chosen table, so the model builds queries against real field names — `linkedinads_get_columns`
- A SQL `SELECT` run against live LinkedIn Ads data — `linkedinads_run_query`
- A `Tables` property in the connection file that narrows what is exposed; blank means everything
A LinkedIn Ads account and a licensed CData JDBC Driver for LinkedIn Ads — trial licences are accepted via `java -jar cdata.jdbc.linkedinads.jar --license`. Java runs the server; Maven builds it with `mvn clean install`, producing `CDataMCP-jar-with-dependencies.jar`. A `.prp` connection file carries `DriverClass` as `cdata.jdbc.linkedinads.LinkedInAdsDriver`, the driver jar path, a tool `Prefix` such as `linkedinads`, and the JDBC connection string from the driver's Connection String utility — OAuth sources authenticate in the browser at that step. MIT licensed.