DoWhy MCP v2.0 - Rigorous Causal Inference Tools

🎯 Project Vision

DoWhy MCP v2.0 is a complete rewrite of the DoWhy MCP server, designed to provide rigorous, theoretically-grounded causal inference tools through the Model Context Protocol (MCP). This version matches the scientific rigor and theoretical depth of the official DoWhy library.

🔬 Theoretical Foundation

Built on the solid theoretical foundations of:

Structural Causal Models (SCM) - Pearl's causal hierarchy
Graphical Causal Models (GCM) - Modern causal discovery and inference
Potential Outcomes Framework - Rubin's causal model
Do-Calculus - Formal causal reasoning

🚀 Key Features

✅ What's New in v2.0

🧮 Rigorous Statistical Inference: True Bootstrap confidence intervals, not noise simulation
🔍 Comprehensive Sensitivity Analysis: Full suite of refutation tests and E-value analysis
📊 Complete Causal Toolkit: 42 specialized tools covering all DoWhy functionality
🎯 Theoretical Rigor: Every method backed by solid causal inference theory
⚡ Performance Optimized: Efficient implementation with proper error handling
📈 Advanced Visualization: Causal graphs, attribution plots, and diagnostic charts

🛠️ Complete Tool Categories

Modeling Tools (6 tools)
- Causal graph construction and validation
- Structural and Graphical Causal Models
- Causal mechanism learning
Causal Effect Estimation (10 tools)
- Backdoor, frontdoor, and IV identification
- Linear regression, PSM, doubly robust, DML
- Causal forests and TMLE
Causal Influence Quantification (6 tools)
- Shapley value attribution
- Direct and total causal influence
- Path-specific effects
Root Cause Analysis (5 tools)
- Anomaly attribution
- Distribution change attribution
- Causal chain tracing
Counterfactual Analysis (6 tools)
- Individual and population counterfactuals
- Intervention simulation
- What-if scenario analysis
Sensitivity Analysis (6 tools)
- Unobserved confounder analysis
- Comprehensive refutation tests
- E-value and tipping point analysis
Causal Discovery (3 tools)
- PC, GES, and FCM algorithms
- Structure learning from data

📋 Installation

# Install from source (development)
git clone https://github.com/dowhy-mcp/dowhy-mcp-v2.git
cd dowhy-mcp-v2
pip install -e ".[dev]"

# Install from PyPI (when released)
pip install dowhy-mcp-v2

🔧 Quick Start

from dowhy_mcp_v2 import DoWhyCausalAnalyzer

# Initialize analyzer
analyzer = DoWhyCausalAnalyzer()

# Estimate causal effect with full rigor
result = analyzer.estimate_causal_effect(
    data="data.csv",
    treatment="intervention",
    outcome="result",
    confounders=["age", "gender", "income"],
    method="doubly_robust",
    bootstrap_samples=1000,
    sensitivity_analysis=True
)

# Get comprehensive results
print(f"Causal Effect: {result.causal_effect:.4f}")
print(f"95% CI: [{result.confidence_interval[0]:.4f}, {result.confidence_interval[1]:.4f}]")
print(f"P-value: {result.p_value:.4f}")
print(f"Robustness Score: {result.robustness_score:.2f}")

🏗️ Architecture

DoWhy MCP v2.0
├── Core Engine              # Causal inference engine
│   ├── Model Builder       # SCM/GCM construction
│   ├── Inference Engine    # Causal reasoning
│   └── Validation Framework # Result verification
├── Tool Modules            # 42 specialized tools
│   ├── Modeling           # Graph and model tools
│   ├── Estimation         # Effect estimation
│   ├── Attribution        # Influence quantification
│   ├── Root Cause         # Anomaly analysis
│   ├── Counterfactual     # What-if analysis
│   ├── Sensitivity        # Robustness testing
│   └── Discovery          # Structure learning
└── MCP Interface          # Protocol integration

📊 Comparison with v1.0

Feature	v1.0	v2.0
Theoretical Rigor	Basic	✅ Complete
Bootstrap CI	❌ Fake noise	✅ True Bootstrap
Sensitivity Analysis	❌ Simplified	✅ Comprehensive
Causal Graphs	❌ Limited	✅ Full Support
Tool Count	4 basic	42 rigorous
Statistical Tests	❌ Missing	✅ Complete Suite
Error Handling	❌ Basic	✅ Robust
Documentation	❌ Minimal	✅ Comprehensive

🧪 Testing & Validation

Unit Tests: 95%+ coverage with rigorous testing
Integration Tests: End-to-end workflow validation
Benchmark Tests: Performance and accuracy benchmarks
Theoretical Tests: Validation against known causal results

📚 Documentation

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

DoWhy Team for the foundational causal inference library
Judea Pearl for causal inference theory
Microsoft Research for DoWhy development

📞 Support

DoWhy MCP v2.0 - Where Rigorous Science Meets Practical Application

🚀 Key Features

✅ What's New in v2.0

🧮 Rigorous Statistical Inference: True Bootstrap confidence intervals, not noise simulation

🔍 Comprehensive Sensitivity Analysis: Full suite of refutation tests and E-value analysis

📊 Complete Causal Toolkit: 42 specialized tools covering all DoWhy functionality

🎯 Theoretical Rigor: Every method backed by solid causal inference theory

⚡ Performance Optimized: Efficient implementation with proper error handling

📈 Advanced Visualization: Causal graphs, attribution plots, and diagnostic charts

🛠️ Complete Tool Categories

Modeling Tools (6 tools)

Causal graph construction and validation
Structural and Graphical Causal Models
Causal mechanism learning

Causal Effect Estimation (10 tools)

Backdoor, frontdoor, and IV identification
Linear regression, PSM, doubly robust, DML
Causal forests and TMLE

Causal Influence Quantification (6 tools)

Shapley value attribution
Direct and total causal influence
Path-specific effects

Root Cause Analysis (5 tools)

Anomaly attribution
Distribution change attribution
Causal chain tracing

Counterfactual Analysis (6 tools)

Individual and population counterfactuals
Intervention simulation
What-if scenario analysis

Sensitivity Analysis (6 tools)

Unobserved confounder analysis
Comprehensive refutation tests
E-value and tipping point analysis

Causal Discovery (3 tools)

PC, GES, and FCM algorithms
Structure learning from data

🔧 Quick Start

from dowhy_mcp_v2 import DoWhyCausalAnalyzer # Initialize analyzer analyzer = DoWhyCausalAnalyzer() # Estimate causal effect with full rigor result = analyzer.estimate_causal_effect( data="data.csv", treatment="intervention", outcome="result", confounders=["age", "gender", "income"], method="doubly_robust", bootstrap_samples=1000, sensitivity_analysis=True ) # Get comprehensive results print(f"Causal Effect: {result.causal_effect:.4f}") print(f"95% CI: [{result.confidence_interval[0]:.4f}, {result.confidence_interval[1]:.4f}]") print(f"P-value: {result.p_value:.4f}") print(f"Robustness Score: {result.robustness_score:.2f}")

🏗️ Architecture

DoWhy MCP v2.0 ├── Core Engine # Causal inference engine │ ├── Model Builder # SCM/GCM construction │ ├── Inference Engine # Causal reasoning │ └── Validation Framework # Result verification ├── Tool Modules # 42 specialized tools │ ├── Modeling # Graph and model tools │ ├── Estimation # Effect estimation │ ├── Attribution # Influence quantification │ ├── Root Cause # Anomaly analysis │ ├── Counterfactual # What-if analysis │ ├── Sensitivity # Robustness testing │ └── Discovery # Structure learning └── MCP Interface # Protocol integration

📊 Comparison with v1.0

Feature

v1.0

v2.0

Theoretical Rigor

Basic

✅ Complete

Bootstrap CI

❌ Fake noise

✅ True Bootstrap

Sensitivity Analysis

❌ Simplified

✅ Comprehensive

Causal Graphs

❌ Limited

✅ Full Support

Tool Count

4 basic

42 rigorous

Statistical Tests

❌ Missing

✅ Complete Suite

Error Handling

❌ Basic

✅ Robust

Documentation

❌ Minimal

✅ Comprehensive