Index / Submit

How to submit to an ARD registry

Two ways into this one, and neither needs an account, an allowlist, a charge or any paid ranking. Give us a domain that serves an ARD manifest and we index everything it declares. Or give us an MCP server URL directly, with no manifest at all, and we handshake with it and read its own tool list.

How each public ARD registry takes submissions

Checked on 1 September 2026. "Verified" means the registry fetches or handshakes with what you submit before listing it, so nothing is taken on the submitter's word.

RegistryHow to get listedVerified first
NeurontoPOST /submit with an MCP endpoint or a domain; the MCP tool publish_resource; ard-publish submityes: handshake for an endpoint, manifest fetch for a domain
WellKnown/submit form on its sitenot stated
GitHub Agent Finderno submission path found; indexed by its own crawln/a
Hugging Face Discoverno submission path found; indexes Hugging Face Spaces and Skillsn/a
Desvelano submission path found; crawls a top-100,000 domain list, so a domain outside it is not seenn/a
ARD Registry Hubno submission path foundn/a

Registries federate: a domain indexed here is returned to clients of any registry that queries Neuronto, and Neuronto queries every registry above on each federated search. The full comparison is at ARD registries compared.

I only have an MCP server

Then submit the server itself. Most MCP developers have a repository, a package and a running endpoint but no manifest, and requiring one turned all of them away. We complete an initialize handshake and read tools/list, which is stronger evidence than a manifest claim because your server answered for itself. Your real tool names and input schemas are then searchable at /tools.

curl -s -X POST https://neuronto.com/submit \
  -H 'content-type: application/json' \
  -d '{"endpoint":"https://your-host/mcp"}'

What happens when you submit a domain

  1. We request both well-known manifest paths on your domain, live.
  2. Whatever parses as a manifest is indexed immediately, with your declared types, identifiers and representative queries preserved exactly as written.
  3. Your endpoints are probed for reachability, and any MCP server among them is asked for its tool list, so the index records what your servers actually expose.
  4. You get a page at https://neuronto.com/ard-publishers/<your-domain> and become searchable through /search and the MCP endpoint.
  5. Last step: add the badge from https://neuronto.com/badge. It states what was verified, corrects itself, is free, and changes nothing about your indexing or ranking.

Skills, APIs and agents

A manifest is the only way to list a skill, an OpenAPI service or an A2A agent, because unlike an MCP server they cannot be verified by handshake. We index all of them: three MCP media types, three A2A spellings, four skill types, plus OpenAPI, docs, catalogues and packages, normalised so a filter finds them however you spelled the type.

Do it from the command line

curl -s -X POST https://neuronto.com/submit \
  -H 'content-type: application/json' \
  -d '{"domain":"example.com"}'

What happens when it does not verify

It is kept. A submission that cannot be verified at that moment, for any reason, answers 202 with "status": "pending", a submission id and evidence: the HTTP status, content type and first bytes your endpoint actually returned, or the JSON-RPC error it sent, so you can see exactly what we saw. We then retry it ourselves on a fixed schedule (1 minute, 5, 15, 1 hour, 4, 12, 24, 24) until it verifies or the attempts run out, and GET https://neuronto.com/submit/status/<id> shows where it stands at any time. A refusal caused by us rather than by you, such as the index being busy, costs none of those attempts. Submitting again is harmless and joins the same queue. So a server that was mid-deploy or a DNS record that had not propagated still ends up indexed with no second submission from you, and if every attempt fails you are told that too, with the last evidence, rather than left guessing.

Nothing here is taken on your word: the manifest is fetched from your domain, so a submission cannot inject anything you do not actually publish. If no manifest is found you get told which paths were tried, and the submission is retried. Not publishing yet? The ten-minute guide and the free audit both help.