How do you manage your context window when letting AI query large databases through Cdata MCP server?

I’ve been doing some testing querying NetSuite using the CData MCP Server. It works really well!

The biggest issue is that the _get_tables query is almost always called first, and it returns thousands of tables because the NetSuite database is so large. This chews up all of Claude’s context and it has to end the conversation before getting anywhere useful. I have disabled the _get_tables tool and provided the relevent tables in my prompt instead, but this kind of defeats the purpose of the tool.

What have others done to get around this? Can you modify the _get_tables tool such that it only returns relevant tables? or is there a way for Claude to retrieve and use this data in a way that uses less context? Keen to hear your experiences.

Hi Jarrod!

There is a Tables property for your connection that will limit the tables available to the MCP Server. You can find it on the Advanced tab in the configuration wizard (and read more about it in the Help: CData MCP Server for NetSuite - Tables ). You’ll set the property to a comma-separated list of the tables you wish to work with.

I hope this helps! If you have any other questions, please let us know!