WEKA-MCP
A machine-learning workbench your model can actually use.
WEKA-MCP puts the WEKA machine-learning workbench behind 46 typed Model Context Protocol tools. A language model can load a dataset, train a model, make predictions, and inspect where it fails, while WEKA remains responsible for the computation. This is also my first contribution to an open-source project.
Why I joined the project
Learning machine learning
Understanding machine learning became a priority when I started working on model-driven products at Harver. I do not plan to become a practitioner, but I want enough foundational knowledge to ask better questions, contribute to discussions, and design more thoughtfully.
Even with a background in Computer Science and Management, I found it difficult to grasp. Learning from Harver’s data scientists has been a steep but valuable curve.
Why WEKA-MCP
At the same time, Ademar began sharing his dissertation work on WEKA-MCP. The open-source project connects WEKA’s machine-learning capabilities with AI tools such as ChatGPT and Claude.
Its focus on non-experts made me say “YES.” It helps students and people with data explore model-building workflows through conversational AI without mastering every technical detail first.
My contribution
I shaped the visual identity, user experience, and front-end implementation of the WEKA-MCP website. The goal was to respect WEKA’s legacy while making its capabilities feel relevant and accessible in the AI era. It was also my first contribution to a GPL-licensed open-source project.
A small behind-the-scenes note: Much of this project was built over weekends in local libraries, often between my day trips to historic cities across the Netherlands. It made the learning process both productive and memorable. 🫶
MCP on one side, WEKA on the other
The project separates the translation layer from the machine-learning workbench. The MCP server forwards typed tool calls to WEKA, so the model can explain results without quietly becoming the place where those results are calculated.
weka-mcp
A TypeScript MCP server that speaks HTTP and stdio, turning each tool call into a WEKA API request.
weka-api
A Java service wrapping WEKA, where training, evaluation, clustering, and diagnostics actually happen.
From data to diagnosis
The toolset follows the way a real analysis unfolds: understand the data, transform it, train a model, evaluate it, and investigate the errors.
Why the boundary matters
Language models are good at explaining results, but they should not invent the statistics they explain. By keeping the computation in WEKA and exposing it through typed tools, every result can be traced back to a real dataset and a real algorithm.
Built on WEKA from the University of Waikato and served over the Model Context Protocol.