What makes ag-ui a protocol?

[JakimLi]

Thanks for the great effort of ag-ui & copilotkit, all the works are very cool, but I have a question about ag-ui as a protocol, which helps me understand what it is, thanks very much.

My questions is what makes ag-ui a protocol?

My understanding is when a protocol is needed, there are at least two parties which might not know each other, or speaks different language or has different rules needs to communicate with each other, for example, MCP is a protocol, because tool provider don't know which client will call it, so it need to follow a protocol, so that the unknown callers are able to call it (the decoupling is also one of the benefits a protocol brings to the parties involved). according to my learnings on the ag-ui site and copilotkit site, what I see is that the ag-ui is a protocol between my backend(agent) and my frontend, both developed by myself, I don't see the value of someone others developing UI for my backend agent. I am confused, what the protocol used for, which parties benefit from it?

I have made some guess, but none of it really convinced me.

1. different agent frameworks, more specific is backend agent(adk, langgraph etc.) framework vs frontend framework(react, angular, vuejs etc.), this brings flexibility and interoperability, but still, the backend and frontend developed by myself, an abstract layer can gain the flexibility and interoperability, why need a protocol?
2. the LLM vs the UI framework, this sounds like MCP(model context protocol) for UI, but still, if I develop the agent also the UI myself, why I need a protocol, I can use SSE / websocket to send the difference and drive UI changes at frontend. unless my agent is going to be used by other client which can follow the protocol and show the UI, and this client don't need to write any UI components/states in advance, which seems not the situation here, even it is the situation, which is generative UI, sounds like a technique, not a protocol.
3. according to the picture on the ag-ui website, it connects the agent <-> user, I don't think the user needs to follow a technical protocol.

Please correct me if my understanding is wrong, and any clear advice to help me understand will be appreciated.

[NathanTarbert]

Hi @JakimLi, apologies for the delay. To your question, what makes AG-UI a protocol?

AG-UI defines a shared contract for agentic UX: event types, state semantics, lifecycle, tool feedback, and human-in-the-loop actions. While you can wire your own backend and frontend with SSE/WebSockets, that only standardizes delivery, not behavior. AG-UI standardizes what is happening and why, so different agents, runtimes, UIs, and tools can interoperate predictably.

What that looks like in practice is:

• It decouples agent logic from UI implementation (like HTTP decouples clients from servers).
• It allows swapping agents, frameworks, or UIs without rewriting UX logic.
• It enables third-party UIs, inspectors, debuggers, copilots, and HITL tools to work generically.
• It creates an ecosystem surface, much like MCP does for tools but focused on user interaction, not model context.

In short, SSE/WebSockets move bytes.
AG-UI standardizes agentic interaction semantics.

[JakimLi]

thank you very much!

[NathanTarbert]

You're welcome, happy to help.