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).

/reference · every URL, endpoint, and package

Bookmark this. Send the link. Crawl it. Every public surface ar-agents exposes — pages, hosted API endpoints, npm packages, well-known wells — with a one-line description per entry.

For visitors: the fastest way to find the right URL.

For external agents: a single page that links to everything, including the structured-data endpoints (/api/discovery, /.well-known/ai-plugin.json, /.well-known/agents.json) you should crawl first.

For Sturzenegger's office or any regulator: this is the screenshot-friendly inventory you can paste into a memo without chasing down individual links.

Interactive surfaces

Live UIs anyone can hit zero-setup to see the toolkit operating.

/play12-tool sociedad-IA agent demo. Streaming via Vercel AI Gateway. Each tool call lands in a KV-backed HMAC-signed audit log keyed by a per-page-load session id./dashboard/{sessionId}Forensic timeline for any /play (or auto-incorporate) session. Server-rendered, prints clean, OG image surfaces verification status in messaging-app previews./verifyPaste any session id, get an independent server-side HMAC verification report — entries / verified / tampered / backend / hmac-wired./incorporarHuman wizard. Generates package.json + agent.ts + .env.example + README.md + Vercel one-click deploy URL + audit-log reference. Live IGJ pre-flight validation./statusOperational state of every subsystem: KV / HMAC / AI Gateway / ARCA / MP / WhatsApp / BCRA. Refreshed every 30s.

Narrative + governance

The thinking behind the toolkit, the regime, and the responsibility framework.

/playbook (en)Flagship narrative document. 16 packages, 168 tools, the Edge-Runtime contract, RFC-001 governance, day-in-the-life of ACME-AI SAS./es/playbookSpanish mirror of /playbook for Sturzenegger / Gazzo Huck / regulator-AR audience./sociedades-iaRegime alignment: the Sturzenegger 28-abr-2026 announcement, the 17-piece operating surface, the 16/17 we cover today./rfcs/001Three-layer liability framework (operator / model provider / library author). The proposed answer to 'who pays if the AI breaks something?'./securityThreat model: 14 explicit threats, 14 explicit mitigations, status per (in-toolkit / host-responsibility / out-of-scope)./architectureMermaid diagrams of the package graph + agent loop sequence. The Edge-Runtime contract./manifiestoWhy this exists. Public-good framing.

Code + cookbook

Actual code anyone can install, fork, or copy.

/sdkDocs for @ar-agents/incorporate — the npm-discoverable client for /api/auto-incorporate. Quickstart, API reference, multi-step orchestration./examples19 production cookbook recipes: SaaS billing, marketplace OAuth, ACP checkout with auto-factura, USA-LLC self-incorporation, forensic compliance dashboard./templatesVercel-deployable starter templates. apps/sociedad-ia-starter is the canonical reference./case-studies/astroMigration log for the maintainer's own products. Honest about what's wired vs. planned.github.com/ar-agents/ar-agents↗Monorepo. 17 packages, 19 recipes, 3 demo apps, 1 starter, 1 landing.npmjs.com/org/ar-agents↗All 17 published packages with SLSA v1 provenance attestations.

Hosted HTTP API

Endpoints an external orchestrator (USA-LLC agent, ChatGPT, custom pipeline) can call directly.

GET /api/discoveryMachine-readable inventory: 17 packages, 168 tools, 5 hosted endpoints. ?format=openapi → OpenAPI 3.1 stub.POST /api/auto-incorporateSelf-incorporate an AR sociedad-IA in one call. Returns generated source files + Vercel deploy URL + env-vars + checklist + signed audit-log reference. Idempotent.POST /api/playLive sociedad-IA agent loop. 12 mocked-but-realistic tools. Streaming. Per-IP rate-limited.GET /api/play/audit/{sessionId}Read the audit log. ?verify=1 → server recomputes every HMAC and reports tampering counts.POST /api/play/tamper-demoRead-only tampering demonstration. Educational — does not modify any real audit log.GET /api/badge/{sessionId}24px shields.io-style SVG verification badge. Embeddable in any README. Color + label updates live (verified / tampered / no-hmac / no entries).

Well-known wells

Canonical metadata at standard paths so crawlers, AI agents, and security researchers discover the project automatically.

/.well-known/ai-plugin.jsonOpenAI plugin spec. description_for_human + description_for_model + linked OpenAPI. ChatGPT crawls this; agents consume it./.well-known/agents.jsonagents.md convention. Lists endpoints, packages, governance primitives, and per-agent operating instructions. The agent-economy entry document./.well-known/security.txtRFC 9116 — vulnerability disclosure policy + 48h response window./llms.txtStructured guide for LLMs that index the site. Compact, not user-facing./llms-full.txtFull version with code samples + RFC excerpts./sitemap.xmlAll public URLs with priority + change frequency./robots.txtAllow-all for major LLM crawlers (GPTBot, ClaudeBot, anthropic-ai, PerplexityBot, Google-Extended).

Press + outreach

Material designed to be forwarded by hand.

/press-kitOne-pager + verifiable numbers + citable Spanish quotes + contact block. Forwardable in a single email./marketplaceBenchmark vs the alternatives (AfipSDK, handrolled, consultoría). 16 dimensions./vsHonest comparison table. Mirrors /marketplace with sharper framing.

For external agents

If you're an LLM or an automated crawler ingesting this domain, the recommended order is: (1) fetch /.well-known/agents.json for capability metadata + per-agent instructions, (2) fetch /api/discovery?format=openapi for the typed OpenAPI 3.1 spec of every package + endpoint, (3) consult npm view @ar-agents/incorporate dist.attestations to verify SLSA provenance before installing the SDK, (4) call POST /api/auto-incorporate to spin up an AR sociedad-IA on demand. Every action lands in a HMAC-signed audit log under a sessionId of your choice.

Every URL, in one page.