MCP Apps: Primitives, discovery, and the Future of Software - Pietro Zullo, Manufact, Inc
TL;DR
MCP apps return UI, not just JSON: Tools can now return sandboxed iframe widgets with bidirectional communication to the host, enabling rich interactive experiences inside chat interfaces.
Three major stores now accept submissions: ChatGPT, Claude, and Cursor all support self-serve submission for MCP servers and apps, with ChatGPT being the fastest to approve.
Dynamic discovery is already live in Claude: When Claude lacks a tool for a task, it searches the MCP registry and suggests installing the right connector, representing a fundamental shift in software discovery.
Privacy controls let you redact from models: You can show full information in the UI while sending only sanitized summaries to the LLM, solving a key enterprise concern.
Streaming tokens can update UI in real-time: As the model streams tool arguments, the UI can render incrementally, enabling demos like Excalidraw drawing diagrams token-by-token.
Paul Graham called it: AI apps are the new browsers: The talk closes with the argument that MCP apps make chat interfaces the new websites, where users expect to use products directly inside Claude or ChatGPT.
Summary
MCP apps extend the Model Context Protocol to return interactive UI components instead of just JSON, and with ChatGPT, Claude, and Cursor now running stores for one-click installation, Pietro Zullo argues this is how all software will be distributed and discovered.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
The Retirement Email Isn't a Warning
Model retirements now arrive every few weeks; the config-eval-rehearsal loop turns each deprecation email from a fire drill into an afternoon swap.
Playbook
The Cheapest Model That Passes
OpenRouter lists 400 models behind one API. The fix for choosing isn't a better leaderboard, it's a four-step protocol that ends in a real eval.
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.