mcp-openvto

MCP SSE server that allows you to do Virtual Try-On through LLM Chat using a back-end db hosted locally or through the cloud (Couchbase) and Replicate APIs.

Attribution

The MCP server uses the following services / libraries.

• cpp-mcp: C++ MCP Client and Server library.
• Eigen C++ library: C++ mathematical linear algebra library.
• Ollama: C++ bindings for Ollama API.
• [Couchbase] (optional)

Requirements

• The .CSV file in the db folder (required if not using Couchbase)
• CSV files alongside embeddings uploaded to Couchbase (local) or Couchbase Capella (Cloud). Works on free tier!
• Ollama
• ... coming soon...

Configuration

Coming soon

Why C++?

I already built 2 MCP servers in Python before and the dependencies and Docker image sizes were too huge. Yes, I could have used an easier example like GoLang but I thought why not!

This project taught me how much I take Python abstractions for granted. 💀 C/C++ humbles you that way.

And yes, doing it in C++ was advantageous.

Example inferences

Basic example

A basic example using Couchbase (Cloud) services.

output.mp4

Regressive inference (Experimental)

An example of a regressive inference using the MCP server with Couchbase (Cloud) services. Regressive here means you can use previous output as the base image to virtual transfer on, e.g. T-Shirt Try On -> Pant Try On using T-Shirt Try On Image.

output.mp4

Meme-ing

Because why not? 😄 Using local CSV vector search.

output.mp4

Contributions

Always welcome to contribute.

License

This project is licensed under the MIT License.