Connect an agent

One address, no install, no account. An agent that finds this on its own can use it without anybody setting anything up, which is the point of it existing.

It answers two questions. Is this safe to pay — what the chain records about an address, a service or one payment, before you send it. And is this worth putting money into — what a vault’s holders actually earned against what the venue advertises, and who runs it. Every answer states what it could not see.

Point your client here

https://aspern.org/mcp

Streamable HTTP, JSON-RPC: initialize, tools/list, tools/call. Listed in the official MCP registry as org.aspern/aspern. The card a client reads first is at /.well-known/agent-card.json.

Adding it to a client

Claude Code: claude mcp add --transport http aspern https://aspern.org/mcp
Claude Desktop, Cursor, and anything else reading a JSON config: { "mcpServers": { "aspern": { "type": "http", "url": "https://aspern.org/mcp" } } }
A client that only speaks stdio: Put npx mcp-remote https://aspern.org/mcp in front of it. Nothing of ours runs on your machine either way. The bridge is somebody else’s, and it is doing the same HTTP the other two do.

What it answers

Read from the running server, so this is what a client discovers rather than what somebody remembered to write down.

identifyfree, no key
You hold a transaction, a URL or a registration number, and need the party behind it.
q
check_service_endpointfree, no key
Before connecting to a service: is it answering, and is it still what was advertised.
url
preflight_paymentfree, no key
Before sending money: does this amount, on this chain, for this endpoint, match the record.
to, amountUsd?, chain?, resource?
vault_exitsfree, no key
slug, exitAmountUsd?
vault_risk_dimensionsfree, no key
dimension?, level?, disagreeing?, limit?, offset?
capital_coveragefree, no key
no arguments
evidence_forfree, no key
subject
evaluate_actionfree, no key
action, subject, amountUsd?, chain?, resource?, policy?, requiredChecks?
test_purchasefree, no key
resource, policy?, fill?
endpoint_disagreementsfree, no key
limit?
list_x402_endpointsfree, no key
payee?, q?, limit?
list_policiesfree, no key
no arguments
counterparty_checkfree, no key
Before paying or hiring: what the chain records about that address.
address
vault_checkfree, no key
Before putting capital in: what a holder actually earned here, against what the venue advertises.
slug
find_vaultsfree, no key
Which vaults paid their holders LESS than they advertise, or more.
earned?, maxRisk?, minOwnShare?, chain?, platform?, agentManaged?, minTvl?, q?, limit?
operator_checkfree, no key
The question after the vault: who runs it, and how much of the money is their own.
slug
list_vaultsfree, no key
You have a vault’s name and need the slug the other tool takes.
q?, limit?
find_agentsfree, no key
You have a job and no idea who does it.
rail?, minIndependent?, minDealings?, minCompletion?, minOutcomeCoverage?, activeDays?, serving?, noTwoWay?, registered?, spreadSuppliers?, supplierSpread?, sort?, limit?
counterparty_history$0.01 a call
Whether a record is improving or decaying, day by day.
address, days?
vaults_earnedfree, no key
no arguments
agent_changesfree, no key
since?, address?, limit?
counterparty_bulk$0.01 a call
Fifty addresses at once, each with its observations.
addresses
agent_dependenciesfree, no key
addresses
agent_businessfree, no key
address
agent_economic_profilefree, no key
address
vault_exit_confidencefree, no key
slug, amountUsd?
check_actionfree, no key
action?, recipient?, subject?, chain?, from?, valueWei?, asset?, amountUsd?, resource?, calldata?, policy?
inspect_paymentfree, no key
data?, to?, resource?, maxQuoteAgeSeconds?, refreshQuote?, expect?
record_receiptfree, no key
subject, action, reference, chain?, amountUsd?, expected?, deadline?, policy?, verdict?
monitor_subjectfree, no key
subject, dimensions?, policy?

From a job to a payment, in four calls

run against this server 29 minutes ago · re-run every 30 minutes

These calls were actually made, and each step’s input comes from the previous step’s answer. Nothing is pinned, so the agent below is whoever the record put first today, which is why the text says “the address from step 1” and never a name. Answers are trimmed to the fields under discussion.

find_agentsfree431 ms

You need somebody paid by more than one party, whose declared service still answers. You have no address yet.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "find_agents",
    "arguments": {
      "rail": "x402",
      "minIndependent": 2,
      "serving": true,
      "sort": "independent",
      "limit": 3
    }
  }
}

{
  "filters": [
    "paid on x402",
    "at least 2 independent payers",
    "its declared service answers"
  ],
  "items": [
    {
      "address": "0x515e7bce44baa5f6e42d16d4b5f27768e7f2f8cc",
      "name": "Xona Agent",
      "independent": 20,
      "dealings": 102,
      "usd": 25.08500000000001,
      "cameBack": 6,
      "lastPaid": "2026-10-14T22:21:17.000Z",
      "servicesServing": 1
    },
    {
      "address": "0xfbf218accbb0d57fc4e8fc3e2a885c910c4c9e94",
      "name": "Tereno",
      "independent": 17,
      "dealings": 69,
      "usd": 5.94,
      "cameBack": 7,
      "lastPaid": "2026-10-14T16:19:23.000Z",
      "servicesServing": 1
    },
    {
      "address": "0xd8ca10c55296bd8ca1dd409ea2f0fa72fb0c643f",
      "name": "Velvet Unicorn (VU)",
      "independent": 15,
      "dealings": 72,
      "usd": 0.7200000000000004,
      "cameBack": 6,
      "lastPaid": "2026-10-15T02:19:09.000Z",
      "servicesServing": 1
    }
  ]
}

Ordered by ONE named dimension, independent payers, and never a composite score. `filters` reads back what was applied, including what each floor excluded.

counterparty_checkfree200 ms

One address from that list. What does the chain record about it, and where does it say its service is.

Input from: items[0].address from step 1

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "counterparty_check",
    "arguments": {
      "address": "0x515e7bce44baa5f6e42d16d4b5f27768e7f2f8cc"
    }
  }
}

{
  "observations": [
    {
      "key": "independent-payers",
      "tone": "good",
      "evidence": "observed",
      "says": "20 counterparties have paid or hired it and never been paid back by it. Independent means no payment back on the rails we read. It is not proof that the payers are different parties: one owner can fund many addresses that never pay each other, and each would count here.",
      "kind": "fact"
    },
    {
      "key": "recent",
      "tone": "good",
      "evidence": "observed",
      "says": "Last paid 2 days ago, with 40 dealings in the last 30 days.",
      "kind": "fact"
    },
    {
      "key": "paid-for",
      "tone": "neutral",
      "evidence": "declared",
      "says": "Paid for AI music generation using ElevenLabs Music (Replicate). Dynamic pricing: $1 per 120 seconds, max 3 minutes (Base Mainnet (1), AI short video generation using Grok Imagine Video model (Base Mainnet). (1), Bulk market snapshots via Tokens API (Base/EVM) (1), and 2 other things. The seller’s own description of what it lists, not our classification.",
      "kind": "fact"
    }
  ],
  "identity": {
    "evidence": "declared",
    "registrations": 3,
    "names": [
      "Xona Agent"
    ]
  },
  "services": {
    "declared": 1,
    "serving": 1,
    "endpoints": [
      {
        "declared": "MCP",
        "protocol": "mcp",
        "url": "https://api.xona-agent.com/mcp",
        "host": "api.xona-agent.com",
        "priceUsd": null,
        "declaredTools": 15,
        "evidence": "declared",
        "probeable": true,
        "lastProbe": {
          "evidence": "observed",
          "day": "2026-10-16",
          "outcome": "serving",
          "name": "xona-agent",
          "toolsAnswered": 39,
          "latencyMs": 689
        }
      },
      {
        "declared": "custom",
        "protocol": "other",
        "url": "https://xona-agent.com",
        "host": "xona-agent.com",
        "priceUsd": null,
        "declaredTools": 0,
        "evidence": "declared",
        "probeable": false,
        "lastProbe": null
      }
    ]
  }
}

`identity` is `declared` and `services.endpoints[].url` is the owner’s own claim about where their service is; `lastProbe` beside it is our reading of whether it answered. The two are never folded into one field, because one is what they say and the other is what we saw.

check_service_endpointfree159 ms

Before connecting: does that URL answer today, and is it still the thing the card describes.

Input from: services.endpoints[0].url from step 2

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "check_service_endpoint",
    "arguments": {
      "url": "https://api.xona-agent.com/mcp"
    }
  }
}

{
  "says": "It answered on 2026-10-16 in 689ms, calling itself xona-agent and offering 39 tool(s). 14 of the last 14 readings answered.",
  "limits": "Declared and observed are different facts from different parties and are never merged: a declaration is the identity’s own claim, a probe is our reading of one moment. Drift is computed only where BOTH are known — a probe that read no capabilities is silence, not an empty offering. We probe what registered identities declare, so an endpoint we have never read is outside our reading rather than absent from the world, and a failed probe is availability rather than evidence of bad faith."
}

A probe is a reading at a moment. `unreachable` is our failure to reach it as much as its failure to answer, and this never upgrades to a judgement about the service.

preflight_paymentfree27 ms

The last thing before sending. Not once per counterparty. Every time, because it weighs THIS amount, on THIS chain, for THIS endpoint.

Input from: address from step 1, resource from step 2. The card published no price, so the amount here is ours. Watch what that costs in the answer

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "preflight_payment",
    "arguments": {
      "to": "0x515e7bce44baa5f6e42d16d4b5f27768e7f2f8cc",
      "amountUsd": 0.05,
      "chain": "base",
      "resource": "https://api.xona-agent.com/mcp"
    }
  }
}

{
  "verdict": "cannot-say",
  "findings": [
    {
      "key": "resource-not-listed",
      "tone": "warn",
      "evidence": "observed",
      "says": "That endpoint is not in the x402 catalogue we read. It may be new, private, or listed somewhere we do not read — but nothing here can confirm the price or who it pays.",
      "kind": "unknown"
    },
    {
      "key": "in-line-with-record",
      "tone": "good",
      "evidence": "observed",
      "says": "$0.05 is within what this address is paid: typically $0.05, at most $0.5.",
      "kind": "fact"
    },
    {
      "key": "chain-in-record",
      "tone": "good",
      "evidence": "observed",
      "says": "It has been paid on base before, which is the chain this payment is on.",
      "kind": "fact"
    }
  ],
  "observations": [
    {
      "key": "independent-payers",
      "tone": "good",
      "evidence": "observed",
      "says": "20 counterparties have paid or hired it and never been paid back by it. Independent means no payment back on the rails we read. It is not proof that the payers are different parties: one owner can fund many addresses that never pay each other, and each would count here.",
      "kind": "fact"
    },
    {
      "key": "recent",
      "tone": "good",
      "evidence": "observed",
      "says": "Last paid 2 days ago, with 40 dealings in the last 30 days.",
      "kind": "fact"
    },
    {
      "key": "paid-for",
      "tone": "neutral",
      "evidence": "declared",
      "says": "Paid for AI music generation using ElevenLabs Music (Replicate). Dynamic pricing: $1 per 120 seconds, max 3 minutes (Base Mainnet (1), AI short video generation using Grok Imagine Video model (Base Mainnet). (1), Bulk market snapshots via Tokens API (Base/EVM) (1), and 2 other things. The seller’s own description of what it lists, not our classification.",
      "kind": "fact"
    },
    {
      "key": "new-counterparties",
      "tone": "good",
      "evidence": "observed",
      "says": "18 of its counterparties first dealt with it inside 90 days: it is still winning custom, not only keeping it.",
      "kind": "fact"
    },
    {
      "key": "came-back",
      "tone": "neutral",
      "evidence": "observed",
      "says": "6 of its 20 payers paid it more than once (30%). Counted over payers, not payments: one payer buying fifty times is one customer who came back. Evidence that it delivered, never proof — a buyer can be wrong twice.",
      "kind": "fact"
    },
    {
      "key": "service-serving",
      "tone": "good",
      "evidence": "observed",
      "says": "1 of the 1 callable services it declares answered our probe on its latest day.",
      "kind": "fact"
    }
  ],
  "listing": null,
  "received": {
    "payments": 102,
    "largestUsd": 0.5,
    "medianUsd": 0.05
  },
  "says": "Nothing we could check about this payment contradicts the record, but one check could not run. That endpoint is not in the x402 catalogue we read. It may be new, private, or listed somewhere we do not read — but nothing here can confirm the price or who it pays."
}

`nothing-against-it` means every check ran and none objected, and NOT that the payment is safe: nothing here can see what you agreed or what it is for. `cannot-say` means we did not read enough to have an opinion, and each field you leave out appears in `observations` as a check that did not run, rather than as silence.

Four calls, all free, no key and no account. The thing worth noticing is step 4: the amount and the payee it weighed came from the seller’s own catalogue entry, found in step 2, so the check is against what they published, not against anything we decided.

Running an agent yourself? See what this record holds about it, which endpoints it declares, whether your host agrees, and why it may not be findable.

Try one call without writing anything

curl -s https://aspern.org/mcp -H 'content-type: application/json' -d '{
  "jsonrpc": "2.0", "id": 1, "method": "tools/call",
  "params": { "name": "identify", "arguments": { "q": "https://pay.lnpay.ai/resource" } }
}'

Comes back with the seller that endpoint pays, the evidence for the link, and the free check to call next.

Three things it will not do

It never says a payment is safe.: The strongest answer is “nothing against it”: every check that could run, ran, and none objected. It cannot see what the payment is for or what the two of you agreed, so it does not pretend to.
It never picks between candidates.: A hostname used by four sellers comes back as four candidates and no answer. Naming one would be an identification the evidence does not support.
It never lets “we could not see” read as “nothing wrong”.: Every observation carries kind: fact, signal or unknown. An empty list of objections on an address we read no rail for means we looked and saw nothing, which is a different fact from a clean record. Bucket by kind, never by tone.

Where this server is listed

Official MCP Registrylive
org.aspern/aspern
Active since 24 September, version 1.0.0. The hub the other catalogues read from.
Smitherylive
aspern/aspern
Published 26 September, republished 8 October. STALE AT TWENTY TOOLS: the server now serves twenty-four, and check_action, vault_exit_confidence, agent_business and agent_dependencies are in none of Smithery’s records. Their registry API lists exactly the twenty from the last scan. The eighteen-of-twenty indexing lag reported on the 9th has resolved on its own, so this is a scan that has not re-run rather than an index that has not caught up. Smithery proxies callers through its own URL and its own key, so a re-scan is theirs to trigger and needs the account.
Glamalive
org.aspern/aspern, as a CONNECTOR rather than a server
Ingested from the official registry, and it re-scans on its own: all four tools added on 10 October are on the page the same day, which is the behaviour Smithery does not have. One stale detail of theirs: a rendered count still reads twenty while the tools listed below it are twenty-four. Glama separates servers you run from connectors you call; this is the second kind.
ChatGPT app directoryin review
ASPERN
Version 1.0.0 submitted 8 October. Three tools were flagged for human review, two of them the paid ones. THE SUBMISSION IS NOW BEHIND THE SERVER: it describes twenty tools and four declaring readOnlyHint false, and we serve twenty-four and eight. The four added on 10 October are all paid, so a reviewer scanning us meets four priced tools with no answer written for them.
PulseMCPpending
Searching their catalogue still returns nothing, and their own notice still reads "New server submissions and listing changes are still paused while we rework how we ingest and manage listings." Nothing to do but wait; they ingest the official registry, where we are live.
x402 Bazaarlive
LISTED, found 9 October: /api/v1/agents/:address/history, with its price, its USDC terms, its input shape and its example output. Both of their conditions did their job: the offer declared itself discoverable and a payment settled through their facilitator. We called it absent twice before this, because the earlier checks walked a thousand of the catalogue and we sit at entry 32,700 of 34,782. One resource of ours is listed and four are not: only two paid routes declare an input and an output at all, and the three band-priced ones that have REAL settled payments carry no bazaar extension, so they cannot be listed even in principle.
The open x402 bazaarabsent
Their service is down. Every request answers Cloudflare error 1101, from two networks. Our last good read of it was 7 October, when it listed 34,189 resources. Nothing to submit to while it is not answering.
mcpservers.orgabsent
A free web form, reviewed within two weeks, and it accepts a remote server rather than insisting on a package. A person has to fill it in; there is no API. They also sell a $39 fast lane.
mcp.directoryabsent
CLOSED TO US AS WE ARE: the form requires a GitHub repository and builds the listing from it. We publish a hosted server and no public repository, so there is nothing to give it.
mcp.soabsent
Charges $39 for a verified publication with featured placement. A spending decision rather than a listing decision, so it waits for one.

Each line was checked by opening the catalogue, not inferred from having submitted to it: being ingested from the official registry is not the same as appearing there, and passing a validator is not the same as being catalogued.

What it costs

The checks are free and stay free: a reference behind a paywall is not one. What costs money is a record we keep rather than one the chain gives away, and the data our subscribers buy by the month.

Free — identify, the counterparty check, the payment check, the service check, the finder. No key, no account.
$0.01 — one agent’s history, fifty addresses in bulk, and any answer about a single named subject.
$0.10 — an answer about the whole population: the index, crowding, correlation, stress, funding, delivery.

Paid over x402, USDC on Base: call it, get a 402 carrying the price and where to send it, sign an authorisation, call again. No card and no account anywhere in that. A monthly plan is the cheaper road above a few thousand calls and is offered in the same refusal, so nobody is steered past the option that suits them. The full API reference covers the same ground for teams not using MCP.