Should we consider adding support for the on-behalf-of token exchange protocol (RFC8693) to MCP server, specifically for agent-to-agent communications? With the growing importance of Agentic AI data access control, passing user access tokens around is not very secure. The RFC supports delegation and can retain both the user identity and the agent identity. This seems to work well for Agentic AI. Would love to hear people’s thoughts and happy to brainstorm on a solution.