statsmodels-mcp

statsmodels-mcp is an MCP server for people writing code with statsmodels. It helps an MCP client choose an estimator, generate starter formulas and code, suggest diagnostics and post-estimation checks, and search the bundled statsmodels docs with a local BM25 index.

Features

• Recommend statsmodels estimators from task characteristics.
• Profile row-oriented sample data and infer likely estimator families.
• Inspect supported model families and compare nearby alternatives.
• Generate Patsy formulas for formula-capable models.
• Generate runnable starter code and fit templates for common statsmodels workflows.
• Rewrite model code and append diagnostics blocks.
• Suggest covariance strategies for heteroskedastic, autocorrelated, or clustered settings.
• Debug fitting failures and explain pasted summary output.
• Suggest diagnostics, robustness checks, and post-estimation analyses.
• Search the vendored docs/*.rst corpus locally with bm25s and extract doc examples.
• Expose MCP tools, resources, prompts, plus HTTP health/readiness routes.

Tools

• list_model_families
• recommend_models
• get_model_details
• profile_dataframe
• infer_model_from_dataframe
• generate_formula_template
• generate_code_example
• fit_model_template
• debug_fit_failure
• explain_summary_table
• add_diagnostics_to_code
• rewrite_model_code
• suggest_covariance_strategy
• suggest_diagnostics
• suggest_post_estimation
• compare_models
• search_docs
• extract_doc_examples
• read_doc_page
• get_docs_for_model
• list_statsmodels_apis

Resources

• statsmodels://server/capabilities
• statsmodels://project/settings
• statsmodels://models/catalog
• statsmodels://docs/manifest
• statsmodels://docs/{doc_path}
• statsmodels://reference/overview

Prompts

• choose statsmodels model
• generate statsmodels workflow
• debug model fit
• interpret regression results

Run

Install dependencies:

uv sync

Run over stdio:

uv run python mcp_server.py --transport stdio

Run over HTTP:

uv run python mcp_server.py --transport http --host 127.0.0.1 --port 8000

Health endpoints:

• GET /healthz
• GET /readyz

Configuration

Important environment variables:

• STATSMODELS_MCP_DOCS_DIR
• STATSMODELS_MCP_DOCS_TOP_K
• STATSMODELS_MCP_TRANSPORT
• STATSMODELS_MCP_HOST
• STATSMODELS_MCP_PORT
• STATSMODELS_MCP_HTTP_HEALTH_PATH
• STATSMODELS_MCP_HTTP_READY_PATH

Example:

STATSMODELS_MCP_DOCS_DIR=./docs \
STATSMODELS_MCP_DOCS_TOP_K=8 \
uv run python mcp_server.py --transport stdio

Testing

just test