EthosPowerAIReference design · Intelligence
GB Battery Site Screening Agent
- Luminus MCP
- Claude
- NESO TEC register
- DNO heat maps
- EA flood map
- Natural England
- ENTSO-E
The problem
A formal connection application costs money and months, and a parcel that fails on flood zone or protected land should never reach one. The public data that would rule it out sits in a dozen portals, each with its own format and its own way of failing.
What I would build
A screening agent runs each candidate parcel through Luminus's MCP tools: terrain, flood zone, protected land, grid proximity, distribution headroom from the five DNOs' published data, queue depth at the nearest grid supply point, a rules-based Gate 2 checklist and, where an ENTSO-E key exists, a revenue sketch from price spreads. Every layer gets pass, warn or fail, and a layer that failed to load is graded unknown rather than clear. A person decides which parcels go forward, and every report says it is not a connection offer.
The logic flow
Click any node for what it does, what it hands on, and where a person is required.
What it returns
Not measured: this is a reference design on someone else's project, not a system I run. One keyless run on 2026-10-17 put a candidate point in Flood Zone 3 in a single call, the kind of finding that should end a site before money is spent; the tools' correctness beyond that run is untested by me.
Why it fits
A storage developer used to pay a consultant for the first look at a parcel, which meant the flood map, the designations, the nearest line and what the network operator's heat map said about the substation down the road. Every one of those layers is public. The developer's own assistant can now fetch all of them before the kettle boils. The trap is to treat that as the old desktop study done faster. Asked for 'the best five sites', an assistant will rank parcels whose grid layer never loaded, and it will state a missing answer in the same tone as a measured one.
The design I would build keeps the speed and refuses the tone. A screening agent runs each candidate through Luminus's own tools (terrain, flood zone, protected land, distance to lines, distribution headroom from the operator's published data, the transmission queue at the nearest grid supply point, and a rules-based Gate 2 checklist) and grades every layer pass, warn or fail, with any layer that failed to load graded unknown rather than clear. On a run on 2026-10-17, one point near Eggborough came back in Flood Zone 3 with 9.4 MW of generation headroom at a primary substation 2.45 km away, while the grid-proximity layer returned nothing because all three of its endpoints failed. The report says each of those things, and a person decides which parcels go forward, and every report carries the line the project's README insists on: this is not a connection offer.
A five-year strategy for a developer or a network company could use the screen, offered as a sketch rather than a plan, as the gate before money, so that nobody pays an application fee, signs a land option or commissions an ecology survey on a parcel the public data already rules out. Year on year the screen becomes the firm's record of why it dropped each site. That record is worth more now that Gate 2, as Luminus's own checklist reads it, asks for a submitted planning application, land rights and a nominated connection point before a queue position. The consultant's hours then go to the parcels that survived, where judgment cannot be fetched. Public data should kill bad sites cheaply.
See it run
Want one of these?
Every part of it is open source and self-hostable. What I bring is the build, the checks around the project's gaps, and the judgement about where your gates should sit.