This is not our API. This repository is an independent, third-party profile of a company's publicly available API surface, maintained by API Evangelist. API Evangelist does not operate, host, resell, or support this company's APIs, and is not affiliated with or endorsed by the company unless stated on the profile.
Where the information came from. Everything here is assembled from material a member of the public can reach with a browser and no credentials — the company's own website, developer portal and documentation, the specifications it publishes for public use (OpenAPI, AsyncAPI, JSON Schema,
apis.json,llms.txtand similar), its public repositories, and its public status, pricing and changelog pages. Nothing here is obtained by breaching a system, defeating an access control, or using credentials of any kind.The rating is an independent assessment. The Kin Score and Agent Readiness rating are independently calculated scores of a company's public API artifacts, produced by API Evangelist against a published rubric. They are not certifications, endorsements, security assessments, or audits, and they score published artifacts — not the quality, safety, or security of the software.
Corrections, re-scores, and removal are free. No partnership, contract, or purchase is required, and you do not need to justify the request.
- Something wrong? Open an issue on this repository, or email info@apievangelist.com.
- Published something new? Ask for a re-score and we will re-run the rating.
- Want the listing taken down? Say so and we will honor it. The profile is reduced to your company name, a factual description, and a link to your own site, and the company is recorded as unrated — never scored zero for having asked.
Response times. Acknowledgement within one business day; removal or restriction within two business days; corrections and re-scores within five business days.
On a security or compliance team? Email info@apievangelist.com with security in the subject line and you will get a person, not a form. We will tell you exactly which public URLs this profile was built from so your team can see the same surface we did, and we will take the listing down on request while you work through it.
Full detail: Where this data comes from
An API Evangelist profile of the Agentic Resource Discovery specification — the third class of repo in this network, alongside API-producing companies and investors. A standard is not a product; it is a coalition with artifacts, so it gets profiled that way: the specifications, the governance, the maturity ladder, and above all the people and companies in the room.
- Site — https://agenticresourcediscovery.org
- Specification — https://agenticresourcediscovery.org/spec/ (v0.9 Draft, Proposal, 28 May 2026)
- Source — https://github.com/ards-project
- Licence — Apache-2.0
- Catalog entry — https://standards.apievangelist.com/store/agentic-resource-discovery/
- The measured population — https://apis.io/ard/
ARD specifies the discovery layer that sits in front of every agentic protocol: the step before
invocation, where a client asks "what is available for this task?" and gets back a ranked set of
MCP servers, agent cards, skills, workflows and APIs. Publishers describe their resources once in an
AI Catalog manifest at /.well-known/ai-catalog.json on their own domain, anchored by a
urn:air: URN derived from that domain. Independent discovery services crawl those manifests and
answer questions over them through a small REST interface whose only mandatory endpoint is
POST /search.
It does not execute anything — MCP, A2A and OpenAPI keep that job. It does not try to be the one registry: it inverts submit-to-a-registry into publish-to-your-own-domain, and expects many competing indexes to federate.
| Path | What it holds |
|---|---|
apis.yml |
Identity — x-type: standards-body, the two contracts the standard publishes, the full property set |
specification/ |
The verbatim specification, the URN naming guide, the AI Catalog spec, and every docs page |
openapi/_original/ |
The ARD Registry API as the standard publishes it — analyse from here, never from a refined copy |
json-schema/ |
The authoritative CDDL grammar and the ai-catalog.json JSON Schema |
governance/ |
The governance read, the ADR ledger, CODEOWNERS, and the observed internal inconsistencies |
conformance/ |
The project's own conformance CLI, its examples, and what it does and does not test |
taxonomy/ |
The standard's own classification — media types, endpoints, discovery mechanisms, federation modes |
people/ |
44 named humans: three authors, three code owners, forty-one acknowledged contributors |
companies/ |
The coalition graph — eleven logo-wall organizations, matched against the network |
leads/ |
Coalition members not yet in the network. There are none, and that is the finding |
adoption/ |
Who publishes a manifest, who merely claims the standard, and what the four registries return |
repositories/, releases/, contributors/, working-groups/ |
The GitHub working surface, and the absences |
Read adoption/ and governance/ for the evidence. In short:
- Nine publishers. Probing
/.well-known/ai-catalog.jsonacross 43,834 domains — every absolute host in the network plus each host's derived apex, 37,751 of them reachable — found nine publishers. None is fully conformant. Four pass the project's own conformance tool. - The authors are behind the adopters. Across 127 domains belonging to the eleven organizations on ARD's contributors wall, two serve a manifest: Hugging Face, and Cisco — whose reference implementation fails the project's own conformance test. Google, Microsoft, GitHub, Databricks, Salesforce, Snowflake, Nvidia, GoDaddy, ServiceNow and AWS publish nothing on their own domains.
- Publishing a manifest does not correlate with being a better API provider. The nine publishers average a 57.5 Kin Score and 44.0 Agent Readiness; the eleven coalition companies average 57.0 and 45.1. Statistically the same population.
- The
urn:ai:publishers are stale, not sloppy. ADR-0009 moved the URN namespace identifier fromaitoairfor validity. Two publishers and one reference registry still emiturn:ai:— they implemented against the earlier draft, and nothing in the ecosystem told them. - Federation is specified, implemented and empty. The only referral any of the four registries returns points at a second index run by the same operator.
- The documentation contradicts the tooling in three places — including the copy-pasteable
template in the publishing guide, which warns under the project's own conformance CLI. All three
are listed in
governance/, with what each one costs. - There is no release train. Zero tags, zero releases; the version lives in a line of markdown, and three artifacts disagree about what it is.
Nothing here scores ARD with the provider Kin Score composite. That rubric measures services; a standards body publishes specifications. Where a score appears in this repo it belongs to a company in the coalition, never to the standard.
No fabrication — never an invented participant, seat, or adoption claim; a miss is recorded as a miss. Provenance on every person, because a name on a title page and a name in an acknowledgements list are different strengths of evidence. Absences are recorded rather than omitted: "defined but empty" and "never defined" are different findings. And the standard's own words are reproduced, not paraphrased — its taxonomy is a primary source about its own intent.
Everything here is read from a public artifact, and every claim carries the URL it came from. Pointing at an artifact is the fastest way to change it — open an issue, or mail info@apievangelist.com.