Auto Causal Inference

This MCP server provides automated causal inference capabilities by combining LLM-guided variable classification with DoWhy statistical analysis on SQLite banking data. Built with Python using FastMCP, LangGraph, and OpenAI's GPT-3.5-turbo, it features a single tool that takes treatment and outcome variables, uses AI to classify other variables as confounders, mediators, effect modifiers, colliders, or instruments, generates causal graphs in DOT format, and executes DoWhy code to estimate Average Treatment Effects using backdoor adjustment with linear regression. The implementation includes both standalone agent and MCP server versions, automatically generates plain-language business summaries of causal effects, and works with predefined banking variables like customer demographics, engagement metrics, and Internet Banking activation, making it valuable for business analysts exploring causal relationships in customer data without requiring deep statistical expertise.

Source

Repository: https://github.com/lethienhoavn/auto-causal-inference