crudecode

The calculator half of Crude Code. The Crude Code MCP server serves US well data from state filings; this package is what an LLM (any LLM with a Python sandbox) runs on that data: the decline-curve forecast echo and the valuation economics. It takes the same two inputs the old server-side tools took and prints the same results; only where the code runs changed.

pip install --break-system-packages crudecode==0.1.0a2
crudecode forecast forecasts.json     # commit decline curves, print the echo
crudecode value params.json           # NPV by status, plus wells_monthly.csv, parameters.csv, README.txt
crudecode doctrine                    # the forecasting method (optional reference)

• forecasts.json is the list of forecast entries (a well or a cohort: wells, oil / gas {qi, di, b}, anchor_month, uptime_factor, struck_months, rationale, analog_cohort). The well facts come from the server's deal_inputs tool, saved as deal_inputs.json. Committed curves are kept in crudecode_run.json; re-asserting a well replaces just that well.
• params.json is the deal terms (interest_type, interest, asset_list, economics_overrides). Flat price deck only; there is no futures strip.
• The field-by-field forms are the get_toolkit tool's description on the server.

Two dependencies (numpy, python-dateutil), no network calls, plain JSON/CSV outputs. Apache-2.0. The tests carry results frozen from the old engine (tests/golden/).