can we add a permission argument for mcp connector? just like readonly/write. for readonly, only query/list/show tools can be exposed to AI agent.

[bladexxx]

This is not an issue. just a idea.
can we add a permission argument for mcp connector? just like readonly/write. for readonly, only query/list/show tools can be exposed to AI agent?

[stephen37]

This one might be interesting for you @jaelgu. I guess it could be possible with RBAC capabilities of Milvus.

I am not sure how we could make that simple with MCP so that we avoid having too many different functions though, maybe adding a Read-only argument for example could be a thing and then we remove the functions that would write. It might be a bit dirty in the code though so not sure.

@bladexxx If you have some examples of other MCP servers doing similar things, I'd be happy to have a look so I can have a better idea.

[jaelgu]

I think this would be better addressed using Milvus built-in security features: Milvus Authentication & RBAC:

1. Enable authentication and RBAC in Milvus.

2. Use the arguments milvus-uri and milvus-token ("user_name:user_password") to start the MCP server.

This ensures that user permissions are properly enforced for reading and writing data in Milvus.

If we simply add permission arguments to the MCP connector, it would only control the visibility of tools within the MCP server rather than enforcing actual access permissions.