Kelvin — power electronics components

io.github.OpenConverters/kelvin · v0.1.0 · ai-and-media

Kelvin — power electronics components is healthy: it completed an MCP handshake in 230 ms and exposes 7 tools.

Read-only parts catalogue for power electronics: search, selection, cross-reference.

Endpoint: https://kelvin.openconverters.com/mcp (streamable-http)

No repository listed · Website · JSON · Add to my inventory

Facts

Users (last 30 days)0
HandshakeHealthy
Auth neededNone
Tool count7
Latency230 ms
Last checked (UTC)2026-10-26T12:00:22.209Z

Health history

Checked (UTC)StatusLatencyError
2026-10-26T12:00:22.209ZHealthy230
2026-10-26T12:00:22.117ZHealthy727

Tools (7)

list_familiesEvery component family in the TAS catalogue, with how many parts it holds and whether it has a deterministic selector. Start here.
describe_familyThe exact field names one family can be filtered, sorted and bucketed on, and the designRequirements keys its selector reads. Call this before search_parts on an unfamiliar family rather than guessing field names.
search_partsParametric search over one family: numeric ranges, categorical values, MPN substring and manufacturer, with optional facet counts. This is the catalogue question — 'what parts do we have that ...' — and needs no converter design.
part_detailsOne part by MPN: its extracted specs and, optionally, the full TAS catalogue record behind them (datasheet URL, mechanical drawing, provenance).
recommend_partsKelvin's deterministic selector: rank real parts against a designRequirements block and return the candidates with the margins that ranked them. Same authority Kirchhoff and Heaviside select through — no LLM judgement, reproducible.
spec_distributionHow one numeric spec is distributed across a filtered slice of a family — bucket counts (log or linear) plus how many parts state the value at all. Answers 'what is available' questions that a top-N list cannot.
cross_referenceFind substitutes for a part by MPN — the deterministic, program-only cross-reference the Kelvin web app runs: pre-gate the catalogue at the ranker's own reject edges, then score every candidate on value, ratings, footprint and mount type. No LLM judgement. An original whose own specs are incomplete