Synthiq

Flagship demo · Front Door MCP

Watch an AI agent become your customer’s fastest path in.

Shopify made every store agent-ready in May 2026. We do it for the businesses that book work instead of selling carts — HVAC, dental, legal, salons. Below is a replayed session between an AI assistant and a real MCP server we built: it quotes a repair from plain language, checks real availability, and stages a booking. The customer never fills a form.

front-door-demo · MCP · JSON-RPC 2.0

A recording of real calls against a working server — not a mockup. Interact with it above, or open it full-screen ↗.

Governance is structural

Agents transact. You stay in control.

Agents read, estimate, and stage — humans confirm. Every booking is created as PENDING and the agent is instructed, in words, to relay that honestly. No payment tools exist in this build by design. This matches the emerging agents.md norm that commitments require human approval.

What the agent can call

  • list_services

    The service catalog with flat prices and durations

  • get_service_quote

    Range estimate from a plain-language description

  • check_availability

    Open arrival windows per date, with ZIP validation

  • get_policies

    Payment, financing, cancellation — quoted from source

  • request_booking

    Stages a PENDING booking — a human confirms

  • capture_lead

    Captures the lead when booking is not the right move

The demo is the pattern, not the product

A production Front Door plugs into your real business.

Live adapters

Google Calendar, Housecall Pro, Jobber, or Square for real availability and bookings, with leads pushed straight to your CRM.

Remote transport + auth

Streamable HTTP with authentication, rate limiting, and per-agent identity — not a laptop over stdio.

Publication layer

agents.md, a curated llms.txt, and an ai-catalog.json entry pointing at the live endpoint — published only once it answers, never before.

Notifications + audit log

SMS or push so the office confirms a pending booking in one tap, and a log of every agent interaction.

We run one ourselves — call our live MCP server ↗