lumerical-mcp v0.4.1

Maintenance release with two agent-ergonomics additions and small fixes.

Added

• save_results — fetch results server-side and write them to a compact .npz, returning only metadata (path + array shapes). Large arrays (E/H over many frequencies and monitors) no longer cross the JSON tool channel and blow the assistant's context; kind:'transmission' stores the Poynting-integrated power spectrum, with optional frequency decimation.
• Monitor apodization control on add_power_monitor and add_scattering_box (apodization None/Full/Start/End, plus center and time width). The cross_section analysis group's child monitors are constructed objects that can't be edited, so add_scattering_box is the settable path for apodized cross-sections.

Fixed

• The MCP handshake reports the package version instead of the mcp library version.
• recipes() resolves common synonyms (e.g. mie → cross_section, sparameters → ports).
• diagnose() recognises "there are no monitors in the simulation".

Install

pip install https://github.com/davydov-nikita/lumerical-mcp/releases/download/v0.4.1/lumerical_mcp-0.4.1-py3-none-any.whl