emovi-mcp

MCP server for the ESRU-EMOVI 2023 social mobility survey (Mexico).

Servidor MCP para la encuesta ESRU-EMOVI 2023 de movilidad social en México.

What is this? / ¿Qué es esto?

emovi-mcp lets AI assistants (Claude, ChatGPT, etc.) query Mexico's most comprehensive social mobility survey through natural language. It exposes weighted statistical computations, intergenerational transition matrices, and variable exploration as MCP tools.

emovi-mcp permite que asistentes de IA (Claude, ChatGPT, etc.) consulten la encuesta de movilidad social más completa de México mediante lenguaje natural. Expone cómputos estadísticos ponderados, matrices de transición intergeneracional y exploración de variables como herramientas MCP.

About ESRU-EMOVI 2023

The ESRU-EMOVI 2023 survey, conducted by the Centro de Estudios Espinosa Yglesias (CEEY), is a nationally representative survey of social mobility in Mexico. It covers 17,843 respondents aged 25-64, with expansion factors representing ~60 million people.

Datasets included:

Installation / Instalación

# Clone the repository
git clone https://github.com/Lalitronico/emovi-mcp.git
cd emovi-mcp

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux/Mac
.venv\Scripts\activate     # Windows

# Install in editable mode with dev dependencies
pip install -e ".[dev]"

# Optional: visualization support
pip install -e ".[dev,viz]"

Or install directly from PyPI:

pip install emovi-mcp
# With visualization support
pip install emovi-mcp[viz]

Prerequisites / Prerrequisitos

• Python >= 3.10
• ESRU-EMOVI 2023 .dta files (obtain from CEEY)

Configuration / Configuración

Set the EMOVI_DATA_DIR environment variable to point to the directory containing the .dta files:

# Linux/Mac
export EMOVI_DATA_DIR="/path/to/Esru Emovi 2023/_extracted/3 BASES DE DATOS/Data"

# Windows (PowerShell)
$env:EMOVI_DATA_DIR = "C:\path\to\Esru Emovi 2023\_extracted\3 BASES DE DATOS\Data"

# Windows (cmd)
set EMOVI_DATA_DIR=C:\path\to\Esru Emovi 2023\_extracted\3 BASES DE DATOS\Data

Usage with Claude Desktop / Uso con Claude Desktop

Add the following to your claude_desktop_config.json:

Windows: %APPDATA%\Claude\claude_desktop_config.json macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "emovi-mcp": {
      "command": "C:/path/to/emovi-mcp/.venv/Scripts/python.exe",
      "args": ["-m", "emovi_mcp"],
      "env": {
        "EMOVI_DATA_DIR": "C:/path/to/3 BASES DE DATOS/Data"
      }
    }
  }
}

On macOS/Linux, replace Scripts/python.exe with bin/python.

Tools / Herramientas

The server exposes 11 MCP tools:

Example queries / Ejemplos de consultas

Once connected, you can ask the AI assistant questions like:

• "¿Cuál es la distribución educativa por sexo?" → Uses tabulate(row_var="educ", col_var="sexo")

• "Muéstrame la matriz de movilidad educativa intergeneracional" → Uses transition_matrix(dimension="education")

• "¿Cuál es el ingreso promedio por región?" → Uses weighted_stats(variable="ingc_pc", by="region_14")

• "Compara la movilidad educativa entre hombres y mujeres" → Uses transition_matrix(dimension="education", by="sexo")

• "¿Qué variables hay sobre educación?" → Uses list_variables(search="educ")

Variable Dictionary / Diccionario de variables

The project includes a pre-built dictionary with 792 variables extracted from the official CEEY documentation and .dta metadata. The dictionary supports searching by name, description, dataset, and section.

To rebuild the dictionary from source data:

python scripts/build_dictionary.py

This requires the Diccionario ESRU EMOVI 2023.xlsx file in the data directory.

Running Tests / Ejecutar pruebas

pytest

All 96 tests cover weighted statistics, transition matrices, mobility indices, Taylor-linearized standard errors, financial inclusion, temporal income comparison, visualization, and variable dictionary functionality using synthetic data (no real microdata needed for tests).

Technical Notes / Notas técnicas

• All statistics are weighted using the factor expansion variable (or fac_inc for the financial inclusion module)
• pyreadstat loads .dta files with apply_value_formats=False to avoid crashes from duplicate municipality labels
• padres_edu is constructed as max(educp, educm) following the CEEY .do file methodology
• Wealth index uses PCA on binary asset indicators (Filmer & Pritchett, 2001), as an alternative to CEEY's MCA approach
• Standard errors use Taylor linearization for ratio estimators under stratified cluster sampling (PSU/strata)
• Mobility indices: Shorrocks M, Prais escape probability, intergenerational Pearson r, corner odds ratios
• STDIO transport: The server communicates via standard input/output, compatible with Claude Desktop and other MCP clients

Project Structure / Estructura del proyecto

emovi-mcp/
├── pyproject.toml
├── README.md
├── LICENSE
├── CONTRIBUTING.md
├── .github/workflows/
│   ├── ci.yml                    # CI: pytest on Python 3.10/3.11/3.12
│   └── publish.yml               # Publish to PyPI on release
├── scripts/
│   ├── build_dictionary.py       # One-time dictionary builder
│   └── validate_ceey.py          # Validate against CEEY 2025 reference
├── validation/
│   └── ceey_reference_values.json # CEEY reference matrices
├── src/emovi_mcp/
│   ├── __init__.py
│   ├── __main__.py               # python -m emovi_mcp
│   ├── main.py                   # FastMCP server entry point
│   ├── config.py                 # Environment, mappings, constants
│   ├── data_loader.py            # Lazy .dta loader with cache
│   ├── dictionary.py             # Variable dictionary (JSON-based)
│   ├── stats_engine.py           # Transition matrices, descriptives
│   ├── data/
│   │   └── dictionary.json       # 792 variables
│   ├── helpers/
│   │   ├── formatting.py         # Markdown formatters for LLM output
│   │   ├── labels.py             # Value label resolution
│   │   ├── mobility_indices.py   # Shorrocks, Prais, odds ratios
│   │   ├── survey_variance.py    # Taylor linearization for SE/CI
│   │   ├── validation.py         # Column + filter validation
│   │   ├── visualization.py      # Heatmaps, Sankey, bar charts
│   │   └── weights.py            # Weighted mean, median, quantile, freq
│   └── tools/
│       ├── __init__.py            # Tool registration (11 tools)
│       ├── compare.py             # compare_groups
│       ├── describe.py            # describe_survey
│       ├── financial.py           # financial_inclusion_summary
│       ├── mobility.py            # transition_matrix
│       ├── stats.py               # weighted_stats
│       ├── subset.py              # filter_data
│       ├── tabulate.py            # tabulate
│       ├── temporal.py            # income_comparison
│       ├── variables.py           # list_variables, variable_detail
│       └── visualize.py           # visualize_mobility
└── tests/
    ├── conftest.py                # Shared fixtures (synthetic data)
    ├── test_dictionary.py
    ├── test_financial.py
    ├── test_mobility.py
    ├── test_mobility_indices.py
    ├── test_stats_engine.py
    ├── test_survey_variance.py
    ├── test_temporal.py
    ├── test_visualization.py
    └── test_wealth_index.py

License / Licencia

MIT

Acknowledgments / Agradecimientos

Survey data: Centro de Estudios Espinosa Yglesias (CEEY) — ESRU-EMOVI 2023.