stats-compass-mcp 0.2.3

MCP server exposing stats-compass-core tools to LLMs like ChatGPT, Claude, and Gemini

stats-compass-mcp

Turn your LLM into a data analyst. 30+ pandas tools via MCP.

Quick Start

pip install stats-compass-mcp

Claude Desktop

stats-compass-mcp install --client claude

VS Code (GitHub Copilot)

stats-compass-mcp install --client vscode

Claude Code (CLI)

claude mcp add stats-compass -- uvx stats-compass-mcp run

Restart your client and start asking questions about your data.

What Can It Do?

Category	Examples
Data Loading	Load CSV/Excel, sample datasets, list DataFrames
Cleaning	Drop nulls, impute, dedupe, handle outliers
Transforms	Filter, groupby, pivot, encode, add columns
EDA	Describe, correlations, hypothesis tests, data quality
Visualization	Histograms, scatter, bar, ROC curves, confusion matrix
ML Workflows	Classification, regression, time series forecasting

Run stats-compass-mcp list-tools to see all available tools.

Loading Files

Local mode: Provide the absolute file path.

You: Load the CSV at /Users/me/Downloads/sales.csv

Remote/HTTP mode: Use the upload feature (see below).

Remote Server Mode

For Docker deployments or multi-client setups:

stats-compass-mcp serve --port 8000

File Uploads

When running remotely, users can upload files via browser:

You: I want to upload a file
AI: Open this link to upload: http://localhost:8000/upload?session_id=abc123

[Upload in browser]

You: I uploaded sales.csv
AI: ✅ Loaded sales.csv (1,000 rows × 8 columns)

Downloading Results

Export DataFrames, plots, and trained models:

You: Save the cleaned data as a CSV
AI: ✅ Saved. Download: http://localhost:8000/exports/.../cleaned_data.csv

Connect Clients to Remote Server

VS Code (native HTTP support):

{
  "servers": {
    "stats-compass": { "url": "http://localhost:8000/mcp" }
  }
}

Claude Desktop (via mcp-proxy):

{
  "mcpServers": {
    "stats-compass": {
      "command": "uvx",
      "args": ["mcp-proxy", "--transport", "streamablehttp", "http://localhost:8000/mcp"]
    }
  }
}

Docker

docker run -p 8000:8000 -e STATS_COMPASS_SERVER_URL=https://your-domain.com stats-compass-mcp

Client Compatibility

Client	Status
Claude Desktop	✅ Recommended
VS Code Copilot	✅ Supported
Claude Code CLI	✅ Supported
Cursor	⚠️ Experimental
GPT / Gemini	⚠️ Partial

Configuration

Variable	Default	Description
STATS_COMPASS_PORT	8000	Server port
STATS_COMPASS_SERVER_URL	http://localhost:8000	Base URL for upload/download links
STATS_COMPASS_MAX_UPLOAD_MB	50	Max upload size

Development

See CONTRIBUTING.md for development setup.

License

MIT