Data Analytics MCP Toolkit

You set up and run a data analytics MCP server that exposes data loading, cleaning, visualization, and machine learning capabilities through an intent-driven pipeline. You interact with it via an MCP client to load data, visualize it, train models, and evaluate results with simple, practical commands.

How to use

Use the MCP client to send an intent and data source to the Data Analytics MCP Toolkit. The server routes your request to the appropriate pipeline—visualization or machine learning—and returns either a chart (as a base64-encoded PNG) or a summary of metrics and model details. You can perform one-shot analyses by describing your goal (for example, show the distribution of a variable, predict a price from features, or cluster data) or follow a step-by-step workflow to load data, clean it, visualize results, split data for training, train models, and evaluate them.

Practical usage patterns include starting with load_data to bring in your dataset, optionally cleaning it, and then choosing visualization or ML paths. The server supports common visualizations like bar, line, scatter, histogram, box, and heatmap plots, and ML options such as linear and logistic regression, k-means clustering, and evaluation endpoints for regression, classification, and clustering.

How to install

Prerequisites: you need Python installed on your system and access to a command line. You may also use a local development environment like an IDE or editor that can run Python processes.

Install the package in editable mode from the project root to ensure changes are picked up without reinstalling.

Run the server using a standard Python invocation or a tool like UV to run it in the background.

cd /path/to/trying_IBM_MCP
pip install -e .
# or
pip install -r requirements.txt

# From project root, with src on path
PYTHONPATH=src python -m data_analytics_mcp.server

uv run --project . python -m data_analytics_mcp.server

# If you installed in editable mode, you can start from the repo root

Cursor MCP configuration

Add the server to Cursor so you can connect to it from your workspace. You will specify the Python interpreter, the module to run, the working directory, and environment variables when needed.

{
  "mcpServers": {
    "data-analytics": {
      "command": "python",
      "args": ["-m", "data_analytics_mcp.server"],
      "cwd": "/path/to/trying_IBM_MCP",
      "env": { "PYTHONPATH": "src" }
    }
  }
}

Usage tips

One-shot usage lets you describe an intent and provide data as a CSV/JSON string or a URL, returning either a chart or ML metrics with a model summary.

Step-by-step usage guides you through a sequence: load_data to obtain a data_id, then clean_data, plot_* commands, train_test_split, train_*, and evaluate_* as needed.

Available tools

load_data

Load data into the server from a CSV/JSON string or a URL.

clean_data

Clean data by dropping missing values and optionally normalizing features.

plot_bar

Create a bar chart and return it as a base64-encoded PNG.

plot_line

Create a line chart and return it as a base64-encoded PNG.

plot_scatter

Create a scatter plot and return it as a base64-encoded PNG.

plot_histogram

Create a histogram and return it as a base64-encoded PNG.

plot_box

Create a box plot and return it as a base64-encoded PNG.

plot_heatmap

Create a heatmap and return it as a base64-encoded PNG.

train_test_split

Split data into training and testing sets for ML workflows.

train_linear_regression

Train a linear regression model.

train_logistic_regression

Train a logistic regression model.

train_kmeans

Train a k-means clustering model.

evaluate_regression

Evaluate regression model performance and metrics.

evaluate_classification

Evaluate classification model performance and metrics.

evaluate_clustering

Evaluate clustering model performance and metrics.

run_analytics

Route an intent and data source to the appropriate pipeline (visualization or ML) and return results.