For agents
Six files tell an agent what the product is, how to call it, and when not to. None needs a key.
Discovery kit
- GET /llms.txtWhat the product is, when not to use it, and where everything is, for agents (llmstxt.org).
- GET /skill.mdThe installable Agent Skills file for the product.
- GET /skills/index.jsonThe skills manifest, listing each skill file.
- GET /mcp/server-cardThe MCP server card (ext-server-card), at the location it recommends.
- GET /.well-known/mcp/server-card.jsonThe MCP server card, at the location spec section 12 names.
- GET /openapi.jsonThis contract, as JSON.
How to call it
From GET /skill.md
- 1If you have a question in words about sanctions status, a travel advisory level, a tariff, or changes since a date, call
ask. Readplanto see how it was understood. - 2If you have a name, call
resolve_entityfirst. For a person, company, vessel, or aircraft, a name alone never selects one entity: pass an identifier (an IMO number, a list's own ID) as a discriminator, or choose a candidate yourself. - 3With entity IDs, call
fetch_factsfor the fields you need, now or withas_of(valid_at,known_at). - 4Keep the
response_idof any answer you rely on. Later,compare_responsesays what changed and why;get_changesfollows the change feed.
When not to use it
- For opinions, forecasts, scores, trading signals, or adviceThe product states what official sources state, nothing more.
- As the only basis for a legal or compliance decisionA name match is a candidate, not a determination, and a negative result covers only the lists and versions it names.
- For a question outside the catalogIt abstains
NOT_COVEREDrather than guess. The catalog covers sanctions lists (OFAC, UK, EU) and travel advisories (US State Department, UK FCDO) where their sources are active. - When you need data newer than the publishers' own release cadenceEvery item says when it was last verified, and strict mode abstains
STALEinstead of serving old data as fresh.