luminus-py 0.3.1

Notebook-friendly Python client for luminus-mcp

luminus-py

A notebook-friendly Python client for luminus-mcp.

This package starts the existing Node MCP server under the hood, calls tools over stdio, and returns Python-native result objects with optional pandas helpers.

Any MCP tool exposed by luminus-mcp is callable directly as a Python method, so the SDK does not need a hand-written wrapper for every tool.

The SDK also includes geospatial helpers for notebook workflows: to_geojson() for lightweight mapping and to_geodataframe() for GeoPandas users.

Roadmap: see ../docs/python-sdk-roadmap.md.

Install

pip install luminus-py[notebook]

For GIS notebook work:

pip install luminus-py[all]

You also need luminus-mcp itself available on your machine, because the Python SDK starts the existing Node MCP server under the hood.

npm install -g luminus-mcp

Current published versions: luminus-mcp@0.3.1 and luminus-py==0.3.1.

API keys

Keyed tools use the same auth model as the Node server. Resolution order is:

1. environment variables like ENTSOE_API_KEY
2. ~/.luminus/keys.json

Example ~/.luminus/keys.json:

{
  "ENTSOE_API_KEY": "...",
  "GIE_API_KEY": "...",
  "FINGRID_API_KEY": "..."
}

Per-notebook overrides are also supported:

lum = Luminus(profile="trader", env={"ENTSOE_API_KEY": "..."})

Quick start

from luminus import Luminus

with Luminus(profile="trader") as lum:
    prices = lum.get_day_ahead_prices(zone="DE")
    df = prices.to_pandas()

    # Any MCP tool can be called directly
    flows = lum.get_cross_border_flows(from_zone="DE", to_zone="NL")
    site = lum.compare_sites(sites=[
        {"name": "A", "lat": 52.1, "lon": 0.1},
        {"name": "B", "lat": 52.2, "lon": 0.2},
    ], country="GB")

    # GIS-friendly exports
    geojson = site.to_geojson(data_key="rankings")

    # Batch several calls into one DataFrame
    multi_zone = lum.call_many_to_pandas(
        "get_day_ahead_prices",
        [{"zone": "DE"}, {"zone": "FR"}, {"zone": "NL"}],
        parallel=True,
    )

    # One-shot export helpers
    prices_df = lum.call_tool_to_pandas("get_day_ahead_prices", {"zone": "DE"})
    rankings_geojson = lum.call_tool_to_geojson("compare_sites", {
        "country": "GB",
        "sites": [
            {"label": "A", "lat": 52.1, "lon": 0.1},
            {"label": "B", "lat": 52.2, "lon": 0.2},
        ],
    }, data_key="rankings")

Notebook demos

Polished notebook demos live in examples/:

• Trader workflow
• GIS siting workflow
• BESS shortlist workflow

Notebook-first helpers

The Python SDK now ships a few opinionated helpers for high-usage analyst flows:

• lum.get_outages_frame(...)
• lum.get_cross_border_flows_many([...])
• lum.get_grid_proximity_substations(...)
• lum.get_grid_proximity_lines(...)
• lum.get_grid_proximity_snapshot(...)
• lum.get_grid_connection_queue_projects(...)
• lum.get_grid_connection_queue_sites(...)
• lum.get_grid_connection_queue_snapshot(...)
• lum.get_distribution_headroom_matches(...)
• lum.get_distribution_headroom_snapshot(...)
• lum.get_grid_connection_intelligence_snapshot(...)
• lum.estimate_site_revenue_frame(...)
• lum.estimate_site_revenue_estimate(...)

get_distribution_headroom_* currently supports operator="SSEN" and operator="NPG" against the public upstreams wired into the MCP server.

Example:

from luminus import (
    DistributionHeadroomSnapshot,
    GridConnectionIntelligenceSnapshot,
    GridProximitySnapshot,
    Luminus,
    SiteRevenueEstimate,
)

with Luminus(profile="gis") as lum:
    outages = lum.get_outages_frame(zone="DE", type="generation")
    flows = lum.get_cross_border_flows_many([("DE", "NL"), ("FR", "DE")])
    substations = lum.get_grid_proximity_substations(lat=52.0, lon=0.1)
    queue = lum.get_grid_connection_queue_projects(connection_site_query="Berkswell")
    headroom = lum.get_distribution_headroom_matches(lat=50.84, lon=-1.08, operator="SSEN")
    revenue = lum.estimate_site_revenue_frame(
        lat=52.0,
        lon=0.1,
        zone="GB",
        technology="bess",
        capacity_mw=20,
    )

    proximity: GridProximitySnapshot = lum.get_grid_proximity_snapshot(lat=52.0, lon=0.1)
    headroom_snapshot: DistributionHeadroomSnapshot = lum.get_distribution_headroom_snapshot(
        lat=50.84,
        lon=-1.08,
        operator="SSEN",
    )
    intelligence: GridConnectionIntelligenceSnapshot = lum.get_grid_connection_intelligence_snapshot(
        lat=50.84,
        lon=-1.08,
        country="GB",
    )
    if intelligence.nged_connection_signal and intelligence.nged_connection_signal.queue_signal:
        project_count = intelligence.nged_connection_signal.queue_signal.summary.matched_projects
    estimate: SiteRevenueEstimate = lum.estimate_site_revenue_estimate(
        lat=52.0,
        lon=0.1,
        zone="GB",
        technology="bess",
    )

Errors and typed models

• Startup failures now raise LuminusStartupError.
• Tool-side configuration failures raise LuminusConfigurationError.
• Tool-side upstream/data-source failures raise LuminusUpstreamError.
• Dynamic whole-surface access still works through LuminusResult, and common GIS/revenue flows also expose opt-in typed models.

Notes

• Use lum.list_tools() to see the live tool surface for the active profile.

• Use lum.describe_tool("tool_name") to inspect the MCP description/schema metadata.

• Use lum.call_many() / lum.call_many_to_pandas() for generic multi-zone or multi-site notebook pulls.

• Use parallel=True on batch helpers when you want the SDK to fan out across multiple MCP subprocesses.

• Use lum.get_day_ahead_prices_many() and lum.get_generation_mix_many() for common analyst workflows.

• Use lum.compare_sites_rankings() together with lum.compare_sites_rankings_geojson() and lum.compare_sites_rankings_geodataframe() for ranked siting output.

• Use the typed snapshots only when they help notebook readability; the raw dynamic MCP surface is still available.

• Notebook demos live in examples/.

• Use to_geojson() for lightweight mapping and to_geodataframe() when GeoPandas is installed.

• Requires luminus-mcp to be available on PATH, unless you pass an explicit command.

• By default the client starts luminus-mcp --profile <profile>.

• For local repo development you can point it at the built server directly:

lum = Luminus(command=["node", r"C:\Users\skf_s\luminus\dist\index.js"], profile="gis")