Preferred way to configure

[celian-garcia]

I'm opening this issue to discuss about the configuration of the MCP server.

Currently it's a command line args, and I understand that it can be convenient for an MCP server.

But going a step further into the authentication, we would like to leverage the Perses config structures for the client. That basically allow any type of authentication, going from the basic auth, to just giving token, passing by our beloved oauth2 client_credentials.
These configs are not manageable with args but with yaml file and env vars. Context: #59

An easy move would be to say, let's move everything to Perses config struct, but I want to be sure that we understand well how an MCP server is configured usually and if having env file + config file is really manageable.

ping @ibakshay @AntoineThebaud

[prakhar29jain]

I am willing to work on this if agreed upon! @ibakshay

[ibakshay]

Great timing on this discussion. I've been working on a refactor branch that uses Cobra + Viper, which makes adopting the Perses config structure rather straightforward.

We could use something like permcp stdio --config perses-mcp-config.yaml for the Perses client?
What do you think? @celian-garcia

If yes, I will merge my PR and @prakhar29jain can go ahead with this implementation.

Example 1: OAuth2 Client Credentials 
# perses-mcp-oauth.yaml
perses_client:
  url: "https://perses.example.com"
  oauth:
    clientID: "mcp-client-id"
    clientSecret: "${OAUTH_CLIENT_SECRET}"  # from env var
    tokenURL: "https://auth.example.com/oauth/token"

[celian-garcia]

Oh wow ok definitely you went a step forward in the industrialization! I won't have time to check further from holidays but it looks far more serious. 💪

[celian-garcia]

Maybe it's wiser to not urge and wait for a four eyes on your PR wdyt ?