Neuronto

MCP is how an agent calls a tool. ARD is how it finds one.

MCP, the Model Context Protocol, is the connection standard many AI clients use to call external tools: an MCP server exposes a list of tools, and a client such as Claude Code can call them. MCP does not answer the earlier question of which server to use in the first place.

That gap is what ARD, Agentic Resource Discovery, is meant to fill. According to Neuronto’s “What is ARD” page, it is an open specification published in June 2026 by a working group that, the site says, includes Google, Microsoft, Hugging Face, AWS and others. The idea is simple: a publisher places a JSON manifest at /.well-known/ard.json on its own domain, describing each resource and the queries it should answer. Registries crawl those manifests and offer a search API. Nobody has to apply to a marketplace to be found. The site is careful to say ARD is a discovery layer, not a runtime: it does not replace MCP or A2A, it just tells an agent what exists. We could not judge from the site alone how widely the specification has been adopted.

Mostly an MCP index, with the rest around the edges

The homepage breaks the index down by resource type, and the split is lopsided. At review time about 123,600 entries were MCP servers, followed by roughly 4,600 skills and 2,900 OpenAPI-described APIs. A2A agents numbered in the low hundreds. So if you are looking for MCP servers, the coverage is broad; for agents in the A2A sense, it is thin.

Most entries are aggregated rather than discovered by Neuronto itself. The site lists its sources openly: Glama, the official MCP Registry, mcp.so, Smithery, Hugging Face and others, plus its own crawl of the well-known discovery paths. It also queries other public ARD registries at the same time (“federation”) and merges their results into one answer.

The publisher tools are the more unusual half

For people who build tools rather than consume them, Neuronto offers a set of utilities around getting listed and being found:

• A manifest generator that builds an ard.json from what a domain already serves, and a publishing guide.
• A conformance test for checking a manifest or registry against the specification.
• A free audit that checks whether agents can discover a domain and which publishers outrank it.
• A visibility check, labelled a preview, that works out likely queries for an MCP server, skill or agent and asks every public registry and the open web where it ranks.
• ARD-Bench, a retrieval benchmark across registries. It is Neuronto’s own benchmark, run on 31 August 2026, comparing itself with rival registries, which readers should weigh accordingly.

There is also a search API and an MCP endpoint, so the registry can be added to an MCP client with one command.

Pricing: free by default, credits past the limits

The pricing page says search, publishing, the console, the data feeds and the API are free without a key and with no time limit, at allowances sized for one person or one agent (60 submissions and 30 audits an hour). Verifying a domain with a DNS record gives a free key with roughly six times those allowances and longer rank history.

Beyond that, usage is paid in prepaid credits, where one credit is $0.001; packs start at $1 for 1,000 credits and are paid in USDC over the x402 payment protocol, not by card. Publisher monitoring, which alerts you by webhook when an endpoint breaks or a ranking drops, is listed at $19 a month per domain with a 14-day trial. The payments feature is a gateway that puts an HTTPS API behind x402 so buyers can pay per call; the page lists 1,000 paid calls a month free, then 1% of each payment. The site states that ranking position is never sold. We list it as freemium.

Arabic support: none, and it matters less here

The site is in English only. The page declares English as its language, we found no Arabic text and no language switcher, and there is no right-to-left layout. Search queries are written in plain language, but the site does not say anything about Arabic queries, so we cannot confirm how well they would match. Since most tool descriptions in the index are in English, searching in English will likely work better.

Who should use it, and who can skip it

Neuronto suits developers building agents who want tools chosen at run time rather than hard-coded, and MCP server or API authors who want to check whether agents can actually find their work. Its honesty is notable: the pricing page openly says real query volume is usually zero and refuses to sell a search-volume figure.

It is not a tool for non-developers. There is no chat assistant here, nothing to generate text or images, and the whole value depends on ARD gaining wider use. The “verified” label only means a server answered with a tool list; the site itself says it is not a trust or safety rating, so you still need to vet any server before connecting an agent to it.