Standards measurements
Each week APIs.io probes the website of every provider in the catalog for the well-known documents that agent and commerce standards define, and publishes what it finds. The counts always carry their denominators, and every week's file is kept, so the series can be compared over time. The data is free to use under CC BY-NC-SA 4.0.
Latest: 2026-11-02
| Providers with a website | 31534 |
| Distinct hosts probed | 28785 |
| Hosts reachable | 27280 |
| Hosts answering 200 on a nonsense path | 2578 |
Serving /.well-known/ucp | 672 |
Serving /.well-known/ucp.json | 425 |
Serving /.well-known/agent-card.json | 418 |
Serving /.well-known/api-catalog | 853 |
Serving /.well-known/agent-descriptions | 3 |
Serving /.well-known/ai-catalog.json | 410 |
Serving /apis.json | 156 |
Serving /.well-known/http-message-signatures-directory | 94 |
Serving /llms.txt | 7872 |
Serving /.well-known/oauth-authorization-server | 2033 |
Serving /.well-known/openid-configuration | 703 |
Serving /.well-known/robots-content-signals | 1748 |
Serving /.well-known/security.txt | 2579 |
Download: latest JSON · latest CSV
Every week
Company MCP servers and the official registry (2026-11-01)
Of the verified providers on APIs.io with a hosted MCP endpoint, how many have that endpoint in the official MCP registry? This is an upper bound on what is missing, and totals only: no list of companies is published until each endpoint has been checked.
| Verified providers with a hosted MCP endpoint | 1662 |
| Matching a registry remote (URL 293, host 25, domain 18) | 336 |
| Matching no registry remote | 1326 (80%) |
| Of those tested (1308): completed an MCP handshake with no credentials, so confirmed servers | 187 |
| Answered 401, which fits an OAuth-protected server but does not prove one | 997 |
| Did not answer as MCP (404, 403, not MCP, unreachable): likely our error, or gone | 124 |
| Active registry servers with a remote, at the time | 35722 |
- Only remotes are matched, so a company listed in the registry by package alone counts as missing.
- 445 unmatched providers have their name somewhere in a registry entry. Some of those are the company and some are third-party wrappers, so they are not subtracted.
- Some endpoints on our own pages are wrong, such as a blog URL recorded as a server. That is our error, and it can inflate the count.
- Provider pages date from 2026-10-27; the registry was read on 2026-11-01.
Method
- Who is probed. The first Website listed on each provider's APIs.io profile, which is the company's own site. API and docs subdomains are not probed here.
- What counts as a hit. A document of the right shape, not just a 200: a
ucpobject for UCP, an agent card'snameplus an interface field for A2A, and alinksetfor RFC 9727api-catalog. Every host also gets a nonsense control path. Thousands of sites answer 200 to any URL, and counting by status code alone would credit them with every standard. - Who runs it. For UCP, the document's own endpoints decide the operator. It is
shopifyorwixwhen they point at the platform, andselfwhen they are on the provider's own domain. - Known positives. Since 2026-11-03, every run first probes a few hosts per document that served it the week before, including Shopify- and Wix-operated stores for UCP. If more than one of a document's hosts reads as a miss, the run is held and not published. This catches the opposite error to the control path: a run that misses documents that are there, as the 2026-10-30 run did for Wix.
- Agent-facing surfaces. Since 2026-11-03 the probe also reads
/agents.mdand/llms-full.txt, which Shopify began serving from many stores in May 2026, and the proposed/.well-known/lar.jsonmanifest with itsLAR:line inrobots.txt. None has a published schema, so a hit means a document of the right kind on a host that does not answer the nonsense path, not a validated one. - What it does not show. A site that blocks automated requests reads as a miss, so these counts are lower bounds. The probe identifies itself as
APIEvangelistBot. - Correction. The 2026-10-30 run counted 1 Wix host serving
/.well-known/ucp; the 2026-11-02 run counts 205, and 204 of those were already in the 10-30 host list. An independent September census had Wix at 50 of 50, so this is most likely a miss in our first run, not a rollout. We have not established why. Compare weeks with care: the catalog also grew by more than 1,100 hosts between these runs.
Generated by scripts/probe_standard_wellknown.py and scripts/build_standard_evidence.py. Each standard's page shows the providers measured for it, for example UCP and A2A.