RefineData MCP Server

RefineData MCP is a Model Context Protocol (MCP) server that provides data preparation, anonymization, and processing tools for AI agents (Claude Code, Cursor, ChatGPT). It is designed to handle big data using Polars and DuckDB, bypassing typical memory constraints of traditional Pandas applications.

Features

• Big Data Analysis: Uses DuckDB to extract statistics from large CSV and Parquet files without fully loading them into memory.
• Web Scraping Pipeline: Automatically extracts and cleans HTML tables from live URLs.
• Live Database Connection: Connects to PostgreSQL and SQLite via DuckDB to query massive databases out-of-core.
• Auto-Visualization: Generates statistical charts (Heatmaps, Boxplots, Scatter) and sends them directly to the LLM UI as Base64 images.
• PII Anonymization: Integrates Microsoft Presidio to automatically mask Personally Identifiable Information (emails, phone numbers).
• Toxicity Filtering: Uses Detoxify (HuggingFace) to filter out toxic text rows.
• Quantitative Simplifier: Calculates percentiles, skewness, and correlation matrices to provide LLMs with statistical summaries instead of raw data arrays.
• Data Processing: Handles missing data imputation and structural cleaning using Polars.

Installation

1. Create a virtual environment and activate it:

python -m venv venv
# On Windows:
.\venv\Scripts\activate
# On Linux/Mac:
source venv/bin/activate

2. Install the required dependencies:

pip install -r requirements.txt

3. Download the NLP model required for PII masking:

python -m spacy download en_core_web_sm

Configuration for MCP Clients

To use this server with an MCP client (such as Claude Code or Cursor), add the following configuration to your client's MCP settings. Replace the absolute path with the location of this repository on your machine.

{
  "mcpServers": {
    "RefineDataMCP": {
      "command": "python",
      "args": ["-m", "refinedata_mcp.server"],
      "cwd": "/absolute/path/to/refinedata-mcp"
    }
  }
}

Tools Provided

The server exposes the following tools to the connected LLM:

• analyze_data: Extracts descriptive statistics using DuckDB.
• process_data: Cleans structural issues and imputes missing values.
• anonymize_data: Masks PII in specified text columns.
• filter_toxic_data: Removes rows exceeding the toxicity threshold.
• simplify_quant_data: Extracts percentiles and correlations for quantitative data.
• scrape_web_data: Extracts data tables from live URLs.
• query_db: Executes SQL queries on PostgreSQL/SQLite databases directly.
• visualize_dataset: Generates charts (scatter, heatmap, boxplot, bar) and returns them as rendering-ready images.

Testing

Run the comprehensive test suite to see it in action:

python test_agent.py