Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)

Agents Decide What the User Sees: How to Hand Them the Interface via MCP Apps or A2UI

Users increasingly talk to AI agents, and the agent itself decides which piece of your interface to show in the chat. That is why two new disciplines have emerged: AX, which covers how an agent works with your service through APIs and MCP, and UAX, which covers how a person sees your service inside the agent. For UAX, MCP Apps delivers a ready-made block in your design, while A2UI (Google) and json-render (Vercel) deliver only a list of elements. Think about how your card will look in someone else's chat.

AI-processed from TG: @tips_ai (Tips AI / IT & AI); edited by Hamidun News
Agents Decide What the User Sees: How to Hand Them the Interface via MCP Apps or A2UI
Source: TG: @tips_ai (Tips AI / IT & AI). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @tips_ai (Tips AI / IT & AI) ↗
Editor-in-Chief
MCP Appsyour branded interface appears in the chat
A2UI and json-renderthe agent decides how to display the card and the button
The author's viewalmost nobody works on UAX yet, even though agents already choose the UI

Picture this: a user doesn't visit your website. Instead, they write to an AI agent: "Find me tickets for Friday." The agent goes through services, compares options, and decides what to show the person in the chat: text, a card with a price, or a "Buy" button.

If your product isn't ready for this kind of intermediation, nothing of your design or brand will survive in the agent's answer. The author of the Tips AI Telegram channel, building on an idea from Twitter, describes this as two new tasks layered on top of regular UI. AX (agent experience) is how easily an agent can work with your product.

UAX (user experience inside the agent) is how a person sees your product inside someone else's agent. For the second task there are three tools: MCP Apps, Google's A2UI, and Vercel's json-render.

AX and UAX: two sides of one problem

In the past, a product was designed for a single "viewer": a human in front of a screen. Now there are two viewers, and the first one is a program.

AX answers the question: can an agent use the service without problems? This covers WebMCP and MCP (protocols through which an agent calls the product's functions), a regular API, and A2A, the protocol for agents to communicate with each other. For a backend developer, this is familiar territory: clear methods, descriptions, predictable responses.

UAX is about what the person ultimately sees. If you gave the agent only raw data, it will turn that data into text. If you gave it an interface or a description of elements, the user will see a product card, a price, and a button in the chat. According to the author, UAX is currently the most interesting direction, and almost nobody works on it, even though agents already decide which part of the interface to show.

Two ways to enter the chat with your own face

The first path is MCP Apps, an extension of the Model Context Protocol. The agent receives a ready-made piece of interface from your server, and the user sees it in the chat just as it appears on your site: with your design and brand. This gives maximum control over appearance. The downside is logical: you maintain one more interface "surface" and are responsible for making it work across different clients.

The second path is A2UI from Google and json-render from Vercel. Here the agent receives not finished markup but a list of elements: a card, a price, a button. How to display them is decided by the agent and its client. The same set can be rendered in a web chat, in a mobile app, or read aloud by voice. The drawback: you give up part of the control, and your brand style depends on how the client interprets the description.

A practical rule follows. If you need recognizability and complex scenarios, such as configuring a plan or choosing a seat in a hall, take MCP Apps. If you need broad client coverage and simple blocks are enough, the declarative approach is simpler and lighter: you pass a short list of elements rather than a whole interface.

Why it matters for business, and what remains open

The author also makes a commercial argument: the most solvent users come through agents, because the agent has already "warmed them up." The logic is clear: a person arrives not from a random search but with a clearly formulated task. But this is the author's opinion, not measured statistics. The post contains no data on conversion or average order value, so the numbers will have to be verified against your own metrics.

There are other limitations. The post is a personal note, not a study, and the specifications themselves are young, so support in specific clients may change. The author cites the primary sources: the MCP Apps page in the protocol documentation, the Google Developers blog post on combining A2UI and MCP Apps, and the json-render repository in vercel-labs. That is where to start.

What a team can do right now:

  • check AX: does the product have an MCP server, or at least a clean API with clear method descriptions;
  • pick one common scenario, such as a product card with a price and a button, and show it in an agent in two ways;
  • compare how the result looks across different clients and how much effort support takes;
  • measure conversion from agent traffic separately from the regular website.

The bottom line is simple: an interface now has two versions, one for the person on your site and one for the person inside someone else's agent. Almost nobody is designing the second one yet, and that is an opportunity to get ahead while competitors are still limited to a regular API.

Story

Key Questions Answered

What are the key details on AX and UAX: two sides of one problem?

In the past, a product was designed for a single "viewer": a human in front of a screen. Now there are two viewers, and the first one is a program. AX answers the question: can an agent use the service without problems? This covers WebMCP and MCP (protocols through which an agent calls the product's functions), a regular API, and A2A, the protocol for agents to communicate with each other. For a backend developer, this is familiar territory: clear methods, descriptions, predictable responses.

What are the key details on Two ways to enter the chat with your own face?

The first path is MCP Apps, an extension of the Model Context Protocol. The agent receives a ready-made piece of interface from your server, and the user sees it in the chat just as it appears on your site: with your design and brand. This gives maximum control over appearance. The downside is logical: you maintain one more interface "surface" and are responsible for making it work across different clients. The second path is A2UI from Google and json-render from Vercel.

Why it matters for business, and what remains open?

The author also makes a commercial argument: the most solvent users come through agents, because the agent has already "warmed them up." The logic is clear: a person arrives not from a random search but with a clearly formulated task. But this is the author's opinion, not measured statistics. The post contains no data on conversion or average order value, so the numbers will have to be verified against your own metrics. There are other limitations.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…