Discovery (live now)
Machine-readable endpoints — no auth required, rate-limited 60/min/IP:
/.well-known/agent.json— generic agent capability manifest/.well-known/ai-plugin.json— ChatGPT-style plugin manifest/openapi.json— OpenAPI 3.1 contract (auto-generated from zod schemas)/sitemap-services.xml— orderable services × locales/llms.txt+/llms-full.txt— LLM-friendly site corpus
Commerce (planned — Stage 2)
POST /api/v1/agent/quote— get immutable price quote (auth)POST /api/v1/agent/order— place order with idempotency-keyPOST /api/v1/agent/reorder— clone last orderPOST /api/v1/agent/subscription— recurring orders
MCP server (planned — Stage 2/3)
- Hosted SSE:
https://mcp.foodtech-consulting.kz/sse - NPM package:
@foodtech/mcp-server(для self-host) - Tools:
list_services,get_quote,create_order,schedule_recurring, etc.
Identity & trust ladder
Каждый agent_principal привязан к responsible human/org:
- unverified → discover + quote only
- email_verified → orders ≤ $100, HIL approval
- domain_verified → orders ≤ $5K, DNS TXT confirmation
- attested → unlimited, org cert + DPA signed
Human-in-the-loop (HIL)
Заказы выше configurable threshold требуют approval owner-человека через email-link (default 24h window). Это решает риск «hallucinating agent заказывает не то».
Payments (planned — Stage 3)
- Stripe Payment Intents — setup-intent на agent_principal
- Google AP2 (Agent Payments Protocol) — mandate-based
- x402 (HTTP 402) + stablecoin — Stage 4 evaluation
Legal
Использование agent-track регулируется отдельным Agent Terms of Service. Owner-человек несёт ответственность за действия зарегистрированных agents.
Roadmap
Полный план в PLAN-AGENTIC-COMMERCE.md (A0 → A5). Текущий уровень: A0 (Discovery foundation).
Contact
Для early integration access: info@foodtech-consulting.kz.