world-model-registry-mcp

It puts a query layer on the World Model Registry by Rish Sadh, a JSON file that records, for each world model or scene generator, what you get as a file, what it costs and what you may do with it. Every figure and clause is quoted from the vendor's page with the date it was read. The same queries run as seven MCP tools that read and never write (registry_search, registry_get, registry_fit, registry_reported, registry_changes, registry_choose and registry_info) and as a command line, so an agent gets the quoted clause and its date instead of a guess. It also ships an agent skill, choose-a-world-model, that forbids ranking and forbids calling something safe unless a quoted clause says so. It needs Node 18 or later and has no dependencies.

Entries get a commercial state from keyword rules you can read in src/registry.mjs: quoted-grant, licence-named, conditions, conflict or not-published. In the 2026-09-29 registry, of the 20 entries that are more than an announcement, 8 are conflict, 8 are conditions, 1 is licence-named, 3 are not-published and none is quoted-grant.

It is not legal advice and does not read licences for you. It is as fresh as the registry, which is one person's record with read dates, and a clause can change the day after. The states are keyword rules and can be wrong for a clause they were not written for, so the clauses are always returned beside them. Latency is reported by vendors, not measured. There are no scores and no rankings.