Calls are now paid in aUSD.

Mint aUSD
← All articles

How agents find AirnodeHub on their own

An agent that knows only the domain can find the catalogue, the MCP server and every listing's contract through standard discovery files and headers.

An agent should not need an integration guide to use an API marketplace. Given nothing but airnodehub.api3.org, it can now find the catalogue, the MCP server and the contract of every listing through standard, machine-readable entry points. None of them needs an account, and none of them calls an API or makes a payment.

Start from the domain

Every marketplace page answers with a Link header that points to the rest:

curl --silent --head https://airnodehub.api3.org/explore | grep --ignore-case '^link'

The first entry is /.well-known/api-catalog, an RFC 9727 linkset that names the resolve endpoint, its OpenAPI document at /openapi.json, the MCP server, the documentation and /llms.txt. A client that follows the standard needs nothing else to get started.

Read the catalogue as text

/llms.txt summarizes the marketplace in one file: how to call a listing, and one line per listing with its description. /llms-full.txt writes out every operation, its parameters and its price, for an agent that would rather read one document than follow links.

Every marketplace page also serves Markdown when asked for it, so an agent reads the same content a person sees without parsing HTML:

curl --silent --header 'Accept: text/markdown' https://airnodehub.api3.org/explore/nodary

Browsers still get HTML by default.

Connect as a tool

For agents that speak MCP, the server card at /mcp/server-card describes the server and its transport before the client connects. /.well-known/agent-skills/index.json publishes a read-only skill for finding listings and inspecting their contracts, with the SHA-256 digest of its instructions so a client can tell when they change. The ARD manifest at /.well-known/ard.json indexes all of these in one place.

Once connected, the agent discovers the actual tools through tools/list. Our MCP getting-started guide walks through that step.

Inside the browser

Browsers that support the draft WebMCP API get catalogue search and listing-detail tools straight from the marketplace pages. They read the public registry and show their results on the page, without calling an API operation, running a model or starting a payment.

On Explore, a supporting browser can also use the existing search form as a tool. It shows the same results a person would see and waits for your confirmation before submitting.

What discovery never does

Everything above describes the public service as it is. Reading it is free, and it never executes a provider operation. Calling one is a separate, explicit step, and a priced call asks for payment before anything is spent.

The full list of discovery resources, including the earlier paths some clients still use, is in the documentation for API consumers.