For agents

1. Discovery

A front door at a known address that says what the site is and where the machine-readable parts live. No crawling to find out.

Here: llms.txt names the site, the key concepts and every link. robots.txt welcomes AI crawlers to the API by name. The API root is a map. Sitemap and schema.org on every page.

2. A twin for every page

The same figures a person sees, as data, with a stable and documented shape. Ask a page for JSON and you get the data.

Here: Every page with an endpoint links to it in the head (rel=alternate, application/json) and in a Link header; every API answer links back to its page. One catalogue generates the router, the OpenAPI file, the docs and the MCP tools.

3. Semantics it can trust

Definitions, units, provenance and freshness attached to every figure, so the agent knows how far to trust a number and how to cite it.

Here: Every response says which Base block it is from and how stale it may be. Every figure is labelled chain, announce or antseedstats, so a fact and our judgement never merge. The definitions behind every tip are in llms-full.txt.

4. Questions, not just lookups

Resolving names to ids, filtering by intent, estimating costs: the agent asks the site what it would ask a person.

Here: 14 MCP tools, one per endpoint, each with the question it answers. Hosted with no key, or run locally from npm. Search, intent filters and a cost estimate are the next phase.

5. Good citizenship, both ways

The site states how its data may be used and measures who uses it; the agent gets limits it can read and attribution it can quote.

Here: Data under CC BY 4.0 with attribution. 60 requests per minute per IP, announced in X-RateLimit headers. Usage counted per endpoint and client family, never per IP.

Licence

Every figure on this site and in the API is published under CC BY 4.0. Use it, quote it, build on it. Attribute it to AntSeedStats (antseedstats.com) and cite the page it comes from.

What we never publish

No personal data. Usage of the API and the MCP server is counted per endpoint and per client family (curl, python, MCP, browser); IPs and raw user agents are never stored.

Limits you can read

60 requests per minute per IP on the API and the MCP server, no key. The budget is in X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset; past it, a 429 with Retry-After. Responses carry Cache-Control and an ETag.

Facts and judgement

Chain is read from Base. Announce is what a seller publishes about itself, taken as is. Antseedstats is a definition of ours, such as the organic view or the trust score. The label travels with the figure.