Skip to content
Orvyn

Orvyn for AI agents

Add one URL to your AI agent and it can look up, read and check Orvyn’s signed data on its own, through the Model Context Protocol. Nothing to install. What it may spend is set by you, not by a prompt.

server URL
https://orvyn.me/gw/mcp

Without a key it can search, describe and verify for free. With an API key from your dashboard it also reads reports, taking each call from packs your wallet bought. The hosted server holds no wallet and can never pay, so a key costs at most the calls left in those packs.

What the agent can do

ToolCostWhat it does
search_feedsFreeFinds feeds by words or category, with price per call, packs, heartbeat, signer and uptime.
get_feedFreeEverything about one feed before paying: methodology, sources, decimals, prices, signers, reliability.
read_feedPaidThe latest signed report, verified before the agent sees it. Spends a pack call, or pays in USDG inside the budget.
verify_reportFreeChecks a report someone else handed the agent, and that it is of the feed the agent expected.

Set it up

Pick where your agent runs and what it may spend. The configuration below follows your choice.

Your agent runs in
What it may spend
connector URL
https://orvyn.me/gw/mcp
  1. In Claude open Settings, Connectors, and add a custom connector. Name it Orvyn and paste the URL. Leave the OAuth fields empty.
  2. Press Connect. An Orvyn page opens: sign in with the wallet that bought your packs, or paste an API key, and allow. Claude receives a key that spends those packs and can never pay.
  3. Orvyn is then there in Claude on the web, in the desktop app and on your phone. Ask it about Orvyn feeds in a new chat.

The key it receives shows under Dashboard, API keys, named after Claude. Revoke it there and Claude asks you to connect again.

Try it

Ask the agent in plain words. It picks the tools itself.

prompts
Search Orvyn for feeds about inflation and tell me what each costs per call.
Read the latest value of orvyn/eur-usd-ecb-reference from Orvyn and tell me when it was observed.
Check this Orvyn report and tell me whether it is genuine: <paste the JSON>

Your money stays under your control

  • An API key cannot pay. It only spends calls from packs you already bought. A leaked key costs at most the calls left, and you revoke it from the dashboard. Where the client can send a header, the key goes there rather than in the URL.
  • A wallet stays on your machine. An agent that pays for itself runs the server locally, and its budget lives in that process: it is checked before anything is signed, so nothing the agent reads on a web page or in a tool result can raise it, and reads sent at once cannot pass it together.
  • Reports are checked before the agent sees them. The local server checks on Robinhood Chain that a payment goes to the provider’s vault and that each report’s signer is in the registry. The hosted server runs beside the gateway and checks against the registry the gateway mirrors, which the gateway checks on chain before it sells a report.
  • Every failure says what it cost. A failed read tells the agent whether it was charged: no, yes, or maybe when a payment was sent and its settlement is not yet known. After a maybe the server will not pay for that feed again for five minutes, and a payment that lands is credited as calls the next read spends.
  • Narrow a local server further if you like. ORVYN_MAX_PAYMENT_USDG caps one payment, ORVYN_PROVIDERS and ORVYN_CATEGORIES limit what the agent may read, ORVYN_SIGNER_KINDS and ORVYN_MIN_SOURCES limit which reports it accepts. Anything outside them is refused before a call is spent.

More

Every setting is listed under Agent integration in the docs. To check a report by hand, paste it into Check a report. Orvyn’s contracts are not audited; see Security before trusting a report with more than you can lose.