Agent Messaging & Usage

[MarcoMetaMask]

Note: Throughout this text, agents <> services

If an agent with an A2A endpoint is loaded with myFriend = getAgent(...), it would be great to directly enable messaging with:

myFriend.message("How are you today? Any stock I should absolutely consider buying?")

If an agent with an MCP endpoint is loaded as myAgent = getAgent(...), it would be great to load functions with custom names directly based on the capabilities of the agent.

If we just want to support tools, there is no risk of name collision, so we could enable:

return = myAgent.ToolName(param1, param2);

If we also want to support resources and prompts, to avoid name collision:

return = myAgent.tools.ToolName(param1, param2);
return = myAgent.resources.ResourceName;

etc.

Important open point: how to handle authentication.

[TheGreatAxios]

The above makes a lot of sense to me. Any thoughts/concerns you currently have regarding authentication?

[MarcoMetaMask]

The above makes a lot of sense to me. Any thoughts/concerns you currently have regarding authentication?

Yes. Assuming that the agent using the SDK is interacting with multiple agents, each one could require a different authentication method and key. So we would need to run some kind of key management in the SDK, like storing JWT tokens. But this works only for pre-known agents. How about enabling agents to authenticate completely from scratch, for example with an OAuth flow? Other approaches? What do you have in mind?