Case study: eight products that AI agents can find, call and pay

The challenge

AI agents now search, compare and call services for people. An agent must know what a service does, how to call it and how to pay, without a form.

Architecture

• One manifest for each site at /.well-known/agent-access.json, with a JSON schema. It lists the website, the OpenAPI description and the MCP server, and points to the price list.
• The manifest follows a public profile, Agent Access Discovery 0.1, with an open licence.
• Each product has an MCP server next to its HTTP API.
• A paid call answers HTTP 402 with the price. An agent pays per call with x402 (USDC on Base) without an account; in Brazil, people buy credit by Pix.
• Each site also publishes llms.txt for readers that are language models.

In production

Rota Nacional, EditalMD, Radar CNPJ, PontoFato, Agentalog, HookPulse, PageAudit and OKF Index publish the manifest.

• Radar CNPJ manifest
• Manifest schema
• EditalMD
• HookPulse
• PageAudit

Numbers and sources

• 8 products publish the manifest, and each lists 3 interfaces: website, OpenAPI and MCP (manifests read on 14 October 2026).
• Per-call payment with x402 from US$ 0.001 (Rota Nacional price table).

Limits

• The profile is a draft (0.1). It describes access and gives no authority.
• The manifests do not implement the Personal Agent Protocol or PACT.
• Payment with x402 needs a wallet with USDC on the Base network.

Hire me for this

• An MCP server and an OpenAPI description for your product, from one contract.
• Per-call billing for agents (x402) next to the checkout you already have.
• A discovery manifest and llms.txt, so that agents find the right entry point.