Hey everyone, dropped an issue on github : https://github.com/archestra-ai/archestra/issues/5679

I'd love to take on this issue myself, however due to time constraints and lack of experience with this stack, i'll probably just slow things down.

hey! welcome to the community 🙂

thanks for opening this - I think it's definitely a valid feature request. so if I understand correctly you have an external application which you'd like to allow consuming :archestra:-defined MCP gateway(s) in, and the users in that app should auth through those MCP gateway(s) on-behalf-of their own :archestra: credentials ("resolve at call time" tool credential type), correct?

can I ask you to try out the branch behind https://github.com/archestra-ai/archestra/pull/5683? I believe this should do what you're after - any feedback appreciated!

also question about your use-case - is your external application also consuming inference through an :archestra: LLM proxy (using oauth client credentials)?

hey @user! Apologies it was nighttime for me when you sent those messages, I see that you have already went ahead and implemented the feature ! I will try it out and let you know. Thank you !

hey @user, I played around with the features, they work great ! Thank you for such a quick and solid implementation.

However, I'd like to clarify an important point in my original issue :

What I had in mind is the ability to lock down access to specific gateways through specific clients via the allowedGatewayIds list, the idea is that users may access some gateways only if their token has been generated by the OAuth flow of the pre-registered client.

Ideal scenario :

• I create a Chat Interface MCP gateway and expose the Github MCP through it, I do not assign any users to this gateway.

• Users cannot access the Chat Interface MCP Gateway.

• I create a Chat Interface client (authorization_code), I assign the Chat Interface MCP Gateway to this client via allowedGatewayIds

• Users who authenticate to Archestra via the Chat Interface client can access the Chat Interface MCP Gateway.

• I may expose a more limited version of the Github MCP through a different gateway, and give broad access to it. But the full features of the MCP Github are gated behind trusted pre-registered clients only.

However, I do understand that this may be a flawed design. This means that a pre-registered client is able to override RBAC and give users access to MCP gateways that they should not have access to based on their permissions. You could even argue that this may be out of :archestra:'s scope.

I am open to discussion and to providing more details if necessary.

do you mind pulling the latest main and trying again?