Show HN: Stop returning raw JSON from MCP servers, build rich inline UIs

AI chat just got mini apps, and the comments are already at war over whether it’s genius or chaos

TLDR: A new tutorial shows how AI chat tools can display small interactive apps inside conversations instead of plain data dumps. Fans think it makes chatbots finally feel useful, while critics say it muddies a simple system and could open the door to sneaky security problems.

A new Towards AI tutorial is pitching a simple but flashy idea: instead of an AI tool spitting out a wall of raw data, it can show a tiny interactive app right inside the chat. The example is a weather card, but the author says the lightbulb moment came from seeing a Spotify widget appear inline in Claude like a mini music player instead of a sad pile of links. In plain English: your chatbot answer could start looking more like a little app you can poke at.

But the real entertainment is in the comment section, where people instantly split into Team “finally, make this usable” and Team “absolutely not, keep the data clean”. One commenter basically said, “Why are we dressing up a data pipe like a toy app store?” Another went even harder, calling the whole thing a security nightmare and comparing the current system to “prompt injection as a service” — which is about as subtle as flipping a table in public.

That clash is the whole vibe here: one side sees a future of slick, friendly AI chats with built-in controls, while the other sees a glitter-covered trapdoor. Even the jokes write themselves — boring old JSON versus shiny inline widgets is giving spreadsheet people fighting iPad people. So yes, the tech is real, but the real spectacle is watching the community argue over whether this is the next big usability win or the moment chatbots became tiny chaos machines.

Key Points

  • The article is a tutorial on using the MCP Apps protocol extension to build interactive interfaces inside AI agent chats.
  • It presents a Python-based weather card as a working example of an inline MCP app.
  • The tutorial contrasts rich inline UI behavior with returning only raw JSON from MCP servers.
  • The article includes a practical use case involving LangGraph debugging.
  • The content is framed as a developer walkthrough for building interactive applications on top of MCP servers.

Hottest takes

"a fully functional, interactive playlist and music player widget spin up inline" — muhammad-shafat
"I think we SHOULD keep returning data" — jun_lung
"prompt injection as a service" — seanhunter
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.