Hugging Face Spaces just became the app store for LLMs with MCP support

Hugging Face Spaces has quietly become the most interesting thing to happen to LLM tools since function calling. With Gradio 5.28.0 now supporting the Model Context Protocol, the platform’s thousands of AI applications are suddenly plug-and-play tools for any MCP-compatible client — Cursor, Claude Code, Cline, you name it.

The mechanics are straightforward. An MCP server exposes a specific capability — image editing, video transcription, whatever the model was built for. You connect that server to your LLM client, and suddenly your model knows how to do something it couldn’t do five minutes ago. Hugging Face calls it “upskilling” your LLM, and the smartphone app analogy they use is actually pretty apt. Your phone can’t edit photos out of the box either.

What makes this genuinely significant is the scale. Hugging Face Spaces is the largest collection of AI applications anywhere. Now every single one of those spaces implemented with Gradio is a potential tool for your LLM. The company is already positioning this as an “MCP App Store,” complete with a browsable directory at hf.co/spaces where you can filter for MCP-compatible apps. Black Forest Labs’ Flux.1 Kontext[dev] image editor is one of the first highlighted examples — describe a change in plain English, and your LLM now knows how to route that request to a model that can actually execute it.

There’s a practical catch worth mentioning. Public spaces can get congested, so users who want reliable performance will need to duplicate spaces to their own accounts. If the space runs on ZeroGPU, that means paying for a PRO account. It’s not a dealbreaker, but it’s the kind of infrastructure friction that separates “cool demo” from “daily driver.” Still, for developers who’ve been wiring together AI capabilities by hand, the idea of browsing a catalog and connecting a tool in seconds is a genuine step forward. The walled gardens never built this. The open ecosystem just did.