Our customers would like to make policy decisions based on which LLM is being used by the client. We are assuming an architecture in which the Authorization Server is separate from the MCP Server (as described here). In this case, the Dynamic Client Registration would be between the IdP and the MCP client, but the MCP server will only get an opaque client_id in the resulting JWT-format access token, which it won't be able to associate with a specific LLM.
I have some thoughts on how this could be addressed, but I would like to know if:
- Is this a concern that others are also seeing?
- Are there any standard mechanisms to achieve this today in MCP?
If this of interest and there isn't a good way to achieve this in MCP, I would like to propose this as an enhancement.
Our customers would like to make policy decisions based on which LLM is being used by the client. We are assuming an architecture in which the Authorization Server is separate from the MCP Server (as described here). In this case, the Dynamic Client Registration would be between the IdP and the MCP client, but the MCP server will only get an opaque client_id in the resulting JWT-format access token, which it won't be able to associate with a specific LLM.
I have some thoughts on how this could be addressed, but I would like to know if:
If this of interest and there isn't a good way to achieve this in MCP, I would like to propose this as an enhancement.