# Neuronto

> Neuronto is a federated Agentic Resource Discovery (ARD) registry and index.
> One search finds the MCP servers, skills, agents and APIs that can do a task,
> across this index and every other public ARD registry at once.

## What is ARD?

ARD stands for Agentic Resource Discovery: an open specification for how AI
agents find the tools, skills, agents and APIs they need, published in June 2026
by a working group including Google, Microsoft, Hugging Face, AWS, Cisco,
GitHub, Nvidia, Salesforce and Snowflake.

An agentic resource is anything an AI client can call to get work done. ARD
answers "what is available for this task?" and then gets out of the way. It is a
discovery layer, not a runtime, and does not replace MCP or A2A.

## What is an ARD registry?

A registry indexes the resource descriptions publishers serve on their own
domains and answers search queries over them. Publishers describe what they
offer at /.well-known/ard.json. Neuronto is an ARD registry and a publisher.

## Index

15412 resources from 6926 publishers, 14522 verified
to respond. Refreshed continuously.

## API (ARD v0.91, no key, no signup)

- POST https://neuronto.com/search    {"query":{"text":"..."},"federation":"auto"}
- POST https://neuronto.com/explore   facet counts over the index
- GET  https://neuronto.com/agents    deterministic paginated listing
- POST https://neuronto.com/mcp       the same search as an MCP tool

## How to be listed (no key, no signup, no allowlist)

Serving a manifest is not being indexed: a registry has to fetch it. Registries
crawl domain lists they chose, so a domain outside one is invisible to it
indefinitely. To be indexed here, send the domain or the endpoint.

- POST https://neuronto.com/submit   {"domain":"example.com"}
  Both well-known paths are fetched from that domain and whatever parses is
  indexed in the same request.
- POST https://neuronto.com/submit   {"endpoint":"https://example.com/mcp"}
  For an MCP server with no manifest at all: the endpoint must complete an
  initialize handshake and its own tools/list is read back.
- MCP tool publish_resource on https://neuronto.com/mcp does the same thing
  from inside a conversation.
- A submission that does not verify at that moment is not dropped: the answer is
  202 with status "pending", a submission id and the evidence of what the endpoint
  returned, and it is retried automatically for about two and a half days.
  GET https://neuronto.com/submit/status/<id> shows where it stands. One call is
  enough; there is no need to resubmit.
- POST https://neuronto.com/audit    {"domain":"example.com"}
  Reports which registries return you, and whether this one can index you now.

Nothing is taken on the submitter's word: everything listed was fetched from
the domain that claims it. Publishing guide: https://neuronto.com/publish

## What is different here

- federation: auto is implemented. The ARD specification makes it the default
  mode: query peer registries, merge, return one set. Neuronto fans out to every
  public ARD registry concurrently and fuses with reciprocal rank fusion,
  reporting which upstreams answered.
- Passes the specification's conformance tool as both registry and publisher
  with zero errors and zero warnings, including GET /agents as a paginated
  object.
- Media types and discovery URN prefixes are normalised on ingest, so a filter
  for MCP servers returns them under all three spellings in circulation.
- Indexed endpoints are probed; ones that do not answer are demoted.
- Relevance scores are semantic only, never a trust or safety rating.

## How to be discovered

Serve an ARD manifest at /.well-known/ard.json on your domain, with
representativeQueries on every entry. The crawler finds it. No submission form,
no allowlist.

## Discovery documents

- https://neuronto.com/.well-known/ard.json
- https://neuronto.com/.well-known/did.json
- https://neuronto.com/robots.txt (carries the Agentmap: directive)