MCPetal

A lightweight Model Context Protocol (MCP) server for the classic Iris dataset. This repository demonstrates how to build an interactive model service with resources, tools, and prompt templates using mcp[cli] and a trained scikit-learn classifier.

Suggested repository names

• IrisMCP
• ContextIris
• MCP-IrisHub
• IrisBridge
• IrisScope

What this project does

This repository exposes an Iris flower classifier through MCP resources and tools. It includes:

• A trained Iris classification model backed by scikit-learn
• MCP resources for model metadata, schema, feature descriptions, and sample inputs
• MCP tools for prediction, explanation, batch prediction, and model training
• MCP prompts for generating analysis and comparison responses

Key features

• model://info — model metadata and version details
• model://schema — input schema for prediction requests
• model://features — feature descriptions for the Iris dataset
• model://classes — class information for each Iris species
• predict(...) — single-sample prediction endpoint
• get_prediction_explanation(...) — explanation and confidence summary
• batch_predict(...) — predict multiple samples from JSON
• train_model() — retrain and save the model
• prompt templates for deeper analysis and comparison

Quick start

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uv run python server.py

Or run with the MCP inspector:

npx @modelcontextprotocol/inspector uv run python server.py

Recommended workflow

1. Activate the virtual environment.
2. Install dependencies.
3. Launch server.py.
4. Use MCP-compatible tools or the inspector to call resources, tools, and prompts.

Files to know

• server.py — MCP resource/tool definitions and server startup
• model.py — Iris classifier wrapper and persistence logic
• model_config.py — dataset schema, feature names, and descriptions
• iris_model.joblib — serialized trained model file
• main.ipynb — interactive exploration and development notebook

Why this repo is useful

This project is a great starting point for any MCP-driven machine learning deployment. It shows how to combine a classic model with structured MCP interfaces, making the model accessible, explainable, and easy to extend.