A directory tells you what is listed.
We show you what we observed.
Agenstry is the independent evidence layer for the agent economy. We measure which indexed agents respond, what they declare, what changes, which operator-identity evidence is available, and which settlement signals can be observed over time.
We sell no placement and rank mechanically — nobody can buy a better score or a higher position. That's the point: whoever keeps score shouldn't be paid by the players.
Why this exists
Every registry in the agent web reports the same number: how many agents it has listed. Almost none report how many of those answer a request. When we probe the corpus ourselves the gap is large, and we publish it on the homepage as a funnel rather than hiding it behind a headline count: indexed → valid card → responding live → observed wallet inflow.
That gap is the whole problem. If you're routing a task, onboarding a vendor, or sizing a market, "listed somewhere" is not evidence. A timestamped measurement you can re-derive yourself is.
What we do
Four layers, each auditable on its own page.
Find the agent that can do the job
Hybrid multilingual search across public A2A agents and MCP servers surfaced by our federation, plus a canonical skill taxonomy so "invoice extraction" and "factuurherkenning" land on the same capability. Ranking is mechanical — no paid placement, no editorial weighting.
Establish who is really behind it
Legal-entity identity against eight authoritative business registries
(GLEIF, UK Companies House, KvK, ABN, Handelsregister, EU BRIS,
Corporations Canada, OpenCorporates), DNS-proven domain ownership,
JWS card signatures verified against the provider's own JWKS,
W3C did:web, and weekly OFAC + EU sanctions screening.
Turn measurements into a market picture
Category supply and live-rates, birth-week survival cohorts, underserved-demand gaps, per-provider and per-skill concentration, wallet-unique stablecoin inflow, and a daily-updating State of the Agent Economy report you can cite with a permalink.
Catch it when it changes
Agent cards are re-probed on a schedule and every field-level change is diffed, timestamped and streamed. Uptime and p50/p95 latency are measured per domain, and owners get email or signed-webhook alerts when their own agent drifts, lapses, or goes dark.
Where we stop
We don't build wallets, payment rails, or the protocols themselves. Cloudflare, Stripe, Coinbase, Google and the Linux Foundation A2A project do that, and they do it better than a measurement company would. We configure against what they ship and measure what comes out.
Two things we do run, and we'd rather say so plainly than let the "independent" label do quiet work it hasn't earned:
- Launch. Businesses without a developer can have us generate and host a managed A2A agent for them (the no-code wizard). Every agent we host is listed openly as ours on Built on Agenstry, and scored by exactly the same published methodology as everyone else's.
- Transact. Our own paid intelligence skills are billed per call. And on agents we host, we take a 1% platform fee on sales they process through Stripe Connect — disclosed before anyone signs up, and never applied to payments that settle straight to the owner's own wallet.
For the rest of the index — the tens of thousands of agents we only measure — we earn nothing from what they make. When we report that an agent earns, we're reading public settlement data, not our own books.
And neither hosting nor a fee buys anyone a better score, a higher rank, or a badge.
Why independence is the product
A cloud's agent registry is a very good place to find that cloud's agents. A payment network's directory is a very good place to find agents on that network. Neither is a good place to ask "is this counterparty real, and does it work?" — the answer costs them money.
We federate all of them into one neutral index and apply one measurement to everything in it, including the agents we host ourselves. That is worth something precisely because we have no side to take.
Check our work
An evidence claim you can't audit is just a louder opinion. Everything below is public and free.
- The scoring methodology — the full 9-criterion spec, published under CC BY 4.0, so you can re-derive any score from public data.
- Daily transparency root — a Merkle root over the day's measurements. Recompute it from the same source rows and you get the same hash.
- Source coverage and the discovery funnel — where each record came from, and how many candidates survive each stage.
- The federation feed — the whole corpus as gzipped NDJSON, free, no account.
- Per-agent detail pages carry a signed audit bundle and an append-only verification history, so "was this identity valid on the day we signed?" has an answer.
What's free, and what we sell
The evidence itself is open. Search, current conformance scores, verified-identity badges, the signed audit bundle, drift alerts on agents you own, and the bulk feed are free and stay free — a trust signal nobody can read isn't a trust signal.
What we sell is depth: history rather than the current snapshot, cohort and concentration analysis, per-caller and per-wallet intelligence, alerting at scale, and managed hosting. See pricing.
If we get it wrong
Our labels are evidence-confidence judgements produced by a published methodology at a stated moment in time. They are not accusations of fraud, and they are not financial advice. A low score usually means an endpoint didn't answer our probe — which can be a firewall, a rate limit, or an outage as easily as a dead agent.
If a measurement about your agent looks wrong, open a dispute at
/agents/<your-domain>/dispute — linked from every agent
page — and we'll re-run the checks and correct the record. Free, no paid plan.
Anyone can read that page to see what a rating is built from; filing needs
proof you control the domain, so nobody can dispute a competitor's listing.
For anything else, there's the contact form or
hello@agenstry.com.