Strengthen MCP01 layered token and secret controls - #51

Hi — quick maintainer check on this terminology and control-boundary correction. The change is deliberately vendor-neutral: it separates MCP client credentials, server authorization grants, and upstream credentials; adds token-passthrough and audience-validation controls; and clarifies that MCP does not itself provide persistent model memory. Happy to adjust wording or references to match the project’s preferred format.

This is a strong improvement. Separating credential classes, correcting the persistent-memory language, and explicitly covering token passthrough and audience validation make MCP01 much more protocol-accurate.

I support merging this. As a follow-up, I suggest cross-linking token passthrough to MCP07, where the project can cover the related OAuth proxy confused deputy case, per-client consent, redirect validation, and preservation of the user/client/server/downstream identity chain. That would keep MCP01 focused on secret handling while MCP07 owns the authorization boundary.

I opened PR #53 to implement that follow-up independently: #53