Use a Gradio MCP server to extend LLMS. Do you want your favorite large language model (LLM) to not just ask questions? What if you could edit images, browse the web, or organize your email inbox?
Well, now you can! I’ll show you in this blog post.
What is the MCP protocol and how it works in the same way as the smartphone apps we are all used to, but with LLM. How to find thousands of MCP servers via the “MCP App Store”. How to add one of these servers to your favorite LLM to grant you new capabilities. Working with Flux.1 using kontext(dev) using the example.
A simple MCP intro
Model Context Protocol (MCP) is an open standard that allows developers to build secure bidirectional connections between LLM and a set of tools. For example, if you create an MCP server that exposes tools that can transcription videos, you can connect LLM clients (such as cursors, claud codes, klein, etc.) to the server. LLM knows how to transcribe videos on request and use this tool.
In short, MCP servers are the standard way to upskill LLM by granting new capabilities. Think of it like a smartphone app. On its own, smartphones cannot edit images, but you can download the app and do this from the App Store. So, is it only if you have an app store for an MCP server? 🤔
Hugging the space in your face: MCP App Store
Hug Facespace is the world’s largest collection of AI applications. Most of these spaces use AI models to perform visualized tasks. for example:
These spaces are implemented in Gradio, an open source Python package for creating web servers with AI. As of version 5.28.0, Gradio Apps supports the MCP protocol.
In other words, hugging the face space is one place where you can find the ability, aka MCP App Store, with thousands of AI-powered capabilities for LLM.
Want to view the App Store? Please visit this link. You can manually filter MCP compatibility at https://hf.co/spaces.
Example: LLM that allows you to edit images
Flux.1 Kontext (dev) is an impressive model that allows you to edit images from a plain text prompt. For example, if you ask them to “dye my hair” and upload their photo, the model will return the photo, but have blue hair!
Let’s plug this model into LLM as an MCP server and edit the image. Follow these steps:
Create a free account by hugging your face. In Settings, click MCP on the left. You may need to scroll down to view the page. Next, scroll to the bottom of the page. You will see a section called Spaces Tool. In the search bar, enter Flux.1-kontext-dev and select a space called Black-Forest-labs/flux.1-kontext-dev. After you click on it, the page should look like this:
This demo uses a cursor, but the MCP client must follow the same steps. (MCP Settings) Scroll back to the top of the page and click the Setup cursor icon in the AI ​​Assistant section. Next, copy that code snippet and place it in your cursor configuration file.
Here, you can start a new chat session with your cursor and request an edit of the image! For now, please note that images must be available via public URLs. You can create a hugging face dataset and save images online.
Using popular public spaces as a tool can mean you have to wait a long time to receive the results. Once you access a space, you can click (Duplicate this space) to create a private version of the space for you. If your space is using “Zerogpu”, you may need to update and replicate it to your Pro account.
Bonus: You can also search for MCP compatible spaces on the Face MCP server to hug! After completing step 4, you can also ask LLM to find space to accomplish a particular task.
Conclusion
In this blog post, we discussed the exciting new features that Model Context Protocol (MCP) brings to large-scale language models. Grade apps, especially those that hug face spaces, are fully MCP-compliant and have seen how to effectively transform space into a vibrant “app store” of LLM tools. By connecting these special MCP servers, LLM can go beyond basic questions and gain powerful new abilities, from image editing to transcription to what you can imagine!
