Ask HN: When Does Your App Need Its Own UI vs. Using It via MCP?
I'm trying to figure out when an app I'm building actually needs its own user interface versus just supplying data via MCP to Claude or ChatGPT. If Claude/ChatGPT becomes the place where the bulk of work gets done in the future—perhaps analogous to a general-purpose computer—then how should app developers build in this world? I'd love to hear HN users' views on this.
I think the answer depends on whether your app's value is in the interaction model itself or in the data/operations it exposes. If the core value is a specialized workflow that benefits from visual feedback, direct manipulation, or complex state, you need a UI. If it's mostly CRUD or data retrieval, MCP is enough.
For me, the UI is still essential for trust and discoverability. Users want to see what's happening, correct mistakes, and explore capabilities. Relying entirely on a chat interface hides your app's features and makes error recovery painful.
We're building both: a thin UI for setup, configuration, and monitoring, and an MCP server for the actual work. The UI is for humans to manage the system; the MCP is for the AI to do the tasks. That split has worked well.
I suspect most apps will end up with a minimal UI for onboarding and billing, plus an MCP endpoint for functionality. The general-purpose assistant becomes the primary interface, and your app becomes a service it calls.
Don't underestimate the cost of building and maintaining a UI. If your users are already in Claude or ChatGPT all day, forcing them to switch to your web app is friction. MCP lets you meet them where they are.