ar-agents is a Mercado Pago Agent Toolkit for the Vercel AI SDK 6. It ships 89 typed tools that an LLM agent can call directly to drive Mercado Pago billing flows: subscriptions, payments, refunds, checkout pro, marketplace OAuth, cuotas (installments), QR in-store, 3DS challenge resolution, point-of-sale devices, webhooks. Sidecar packages cover AFIP/ARCA, WhatsApp Business Cloud, banking (CBU/CVU + BCRA), and shipping (Andreani/OCA/Correo Argentino).

How is this different from the official mercadopago SDK?

The official mercadopago SDK is a thin REST client. It does not ship Vercel AI SDK tool schemas, does not implement webhook HMAC verification with replay protection, does not run on Edge Runtime (Node-only), and does not gate irreversible operations. ar-agents adds all of that on top of the underlying API: 89 typed tools, deterministic idempotency keys derived from inputs, programmatic human-in-the-loop on refund/cancel/delete, npm provenance attestation, Vercel KV adapters via subpath, OpenTelemetry instrumentation. You can use both packages in the same project; ar-agents wraps the underlying API directly without depending on the official SDK.

Does ar-agents work on Edge Runtime?

Yes. The whole package is Web Crypto-based with no node:crypto dependency, so it runs on Vercel Edge Functions, Cloudflare Workers, Deno, and any other V8-isolate runtime. Webhook signature verification, HMAC, and idempotency-key generation all use the Web Crypto API.

What is HITL (human-in-the-loop) in ar-agents?

Eight tools mutate state irreversibly (refund_payment, cancel_subscription, delete_customer_card, etc.). The toolkit accepts a requireConfirmation callback that gates each invocation: the tool function literally will not execute until your callback returns true. This is a programmatic gate, not just an LLM instruction. You can show the user a UI and wait for their explicit approval before any irreversible operation runs.

How does idempotency work?

Every POST request gets an auto-generated idempotency key. For LLM-driven retries, four mutating tools (create_payment, create_subscription, create_payment_preference, refund_payment) use a deterministic key derived from a SHA-256 hash of the meaningful inputs (external_reference, amount, payment_method, etc.). Same inputs produce the same key, so retries return the existing resource instead of double-charging the customer.

Yes. MIT license. No paid tier, no telemetry phone-home, no usage caps. The package is published to npm under the @ar-agents scope with SLSA v1 provenance attestations.

What about AFIP, WhatsApp, banking, shipping?

Sidecar packages cover the rest of the Argentine business stack: @ar-agents/identity (CUIT/CUIL validation + AFIP/ARCA padron lookup with monotributo category and IVA condition), @ar-agents/facturacion (AFIP/ARCA factura electronica via WSFE), @ar-agents/whatsapp (WhatsApp Business Cloud API with HMAC webhook verify and AR phone normalizer), @ar-agents/banking (CBU/CVU validation + BCRA Central de Deudores), @ar-agents/shipping (Andreani, OCA, Correo Argentino), @ar-agents/identity-attest (HMAC-signed verification orchestrator). Each ships independently to npm.

Is there a Model Context Protocol (MCP) server?

Yes. @ar-agents/mcp bundles all 7 ar-agents packages into a single MCP server compatible with Claude Desktop, Cursor, Codeium, Continue, Cline, or any MCP host. Auto-detects which packages to enable from environment variables. Listed on Glama (glama.ai/mcp/servers/ar-agents/ar-agents) and the official MCP Registry (io.github.ar-agents/mcp).

ar-agents, open infrastructure for AI corporations in Argentina

If Argentina is going to host 500 million AI agents paying taxes here, someone has to write the code that connects them to the state. Let it be open source. Let it be good. Let it be here.

The moment

In April 2026, Argentina's Minister of Deregulation Federico Sturzenegger announced his plan to make Argentina the first country with a legal regime for AI corporations: companies with no humans, only code that decides, operates, and pays taxes. His projection: 50 million inhabitants and 500 million AI agents incorporated here.

The same week, Peter Thielbought a mansion in Buenos Aires and started conversations with the government about Palantir as an intelligence contractor. The narrative of “Argentina = AI jurisdiction” is being written in real time, from the top.

What's missing is the technical layer. Sturzenegger's sociedades IA don't yet have code that makes them exist: there are no libraries for an agent to register a CUIT, issue electronic invoices, pay monotributo, open a MercadoPago account, send a WhatsApp. Whoever builds them first defines the standard.

The thesis

ar-agents is the bet that this infrastructure has to be open, civilian, production-grade, and native to the frameworks already winning (Vercel AI SDK 6, Model Context Protocol, agents.md). No SaaS, no state contracts, no consulting: the technical primitives underneath everything built this cycle.

What's already published on npmjs.com/org/ar-agents:

@ar-agents/mercadopago         89 tools, pagos, suscripciones, marketplace
@ar-agents/mercadolibre        14 tools, items, órdenes, claims, shipments, ML marketplace
@ar-agents/identity             CUIT/CUIL + AFIP/ARCA padrón
@ar-agents/identity-attest      RENAPER-bypass via WhatsApp/email/Auth0
@ar-agents/whatsapp             WhatsApp Business Cloud + AR phone normalizer
@ar-agents/banking              CBU/CVU + BCRA Central de Deudores + Variables
@ar-agents/facturacion          AFIP/ARCA factura electrónica WSFE
@ar-agents/shipping             Andreani / OCA / Correo Argentino
@ar-agents/mi-argentina         OIDC del gobierno argentino, login con CUIL
@ar-agents/boletin-oficial      Firehose estructurado del BO con suscripciones
@ar-agents/igj                  IGJ datos abiertos, sociedades, autoridades, balances
@ar-agents/firma-digital        Verificación Ley 25.506 / ONTI, certs y CMS
@ar-agents/gde-tad              GDE/TAD, Domicilio Electrónico + IGJ pre-flight
@ar-agents/ap2                  Agent Payments Protocol, mandatos SD-JWT VC
@ar-agents/agentic-commerce-bridge   ACP bridge ChatGPT/Claude/Gemini → ML + MP
@ar-agents/incorporate          Client TS para auto-incorporación programática
@ar-agents/mcp                  MCP server unificado para todos los anteriores

Design decisions

Web Crypto, not Node. Everything runs on Edge Runtime, Cloudflare Workers, Deno. If your agent lives in serverless, ar-agents works there.

Adapter pattern by default. Every package ships with an UnconfiguredAdapter that doesn't crash, it returns setup instructions. Tools are always safe to call.

AGENTS.md per package. Following the agents.md convention, each package ships instructions the LLM reads at runtime: when to use, when not, return shape, side effects, latencies.

Deterministic idempotency. Mutating tools derive their key from a hash of the inputs. A retrying LLM doesn't double-charge.

Programmatic HITL. Irreversible operations (refund, cancel, delete) require explicit confirmation BEFORE execution. It's not a model instruction, it's a code gate.

What ar-agents does NOT do

Does not sell to intelligence services. Does not participate in contracts with state intelligence or security agencies. The infrastructure is explicitly civilian, commercial, open (developer-OSS). If you need Palantir Gotham, there are other places.

How to contribute

The repo is github.com/ar-agents/ar-agents. Issues open, PRs welcome, conventional commits, MIT license, npm provenance. If you build a missing package, open an issue before starting, I'll coordinate scope to avoid duplicate effort.

Open letter to the Minister

On May 9, 2026 I published an open letter to Minister Sturzenegger at /to-the-minister. It explains what may be useful to him from the stack, what I'm not asking for (contracts, subsidies, meetings), and the AAIF working group proposal to make the technical profile of AI corporations a neutral international standard. CC0, copy, translate, cite.

Naza, May 2026

Open infrastructure for the Argentine agent jurisdiction.