Agent Card, JSON-LD, and MCP: A Practical Stack for Merchants

Local merchants do not need to become API companies overnight. They do need a small, coherent stack so ChatGPT, Perplexity, Gemini, DeepSeek, and future assistants can find accurate facts, verify identity, and take a next step. The jargon—Agent Card, JSON-LD, MCP—sounds engineering-heavy. Operationally, it maps to four publish surfaces you can finish in weeks, not quarters.

This guide explains each layer, what “good enough” looks like for a restaurant, clinic, or B2B supplier, and how Trusgent wires them together.

The Stack at a Glance

| Layer | Human URL | Machine surface | Job |

| --- | --- | --- | --- |

| Brand Page | `/b/{handle}` | Facade JSON | Hours, menu, booking / WhatsApp actions |

| Trust Passport / Agent Card | `/t/{trusgentId}` | Agent Card JSON (`/api/v1/t/...`) | Capabilities, trust score, verification level |

| Official provenance | `/r/{registryId}` | Registry fields | Licence / CR anchors |

| Tool access | Developer docs | MCP tools + public APIs | `search_agents`, `get_agent`, `verify_agent`, facade tools |

JSON-LD on your own website remains useful for classic SEO/AEO. Think of Trusgent as the canonical Trust Layer mirror that agents can query even when your CMS is messy.

Layer 1 — Brand Page (`/b`): Actions Humans and Agents Share

Publish:

•Opening hours with timezone honesty

•Services or menu items with clear names (avoid cryptic SKUs)

•Primary contact path (phone, WhatsApp, booking)

•Photos that match the real venue

AAO test: Can an assistant truthfully say “Open until 22:00; book via this page” without inventing a form?

Misalignment between `/b` and Google Business Profile is the #1 avoidable trust leak. Align them weekly.

Layer 2 — Agent Card (`/t` + JSON): The Shortlist Object

The Agent Card is the structured résumé assistants compare when building a shortlist:

•Display name and locale variants

•Service regions and categories

•Verification level and trust signals

•Links to Brand Page and registry provenance

GEO test: If two merchants offer “Central lunch,” does your card carry enough disambiguating facts (cuisine, licence class, party size, price band) to survive recombination?

Directory listing ([shortlist article](/blog/agent-directory-ai-shortlist)) further increases discoverability for agents that prefer curated inventories over raw web scrape.

Layer 3 — Registry Provenance (`/r`): Identity Anchors

Claim the official Hong Kong record when available ([claim guide](/blog/how-to-claim-hong-kong-registry-listing-on-trusgent)). Link it to your agent so verification is not a marketing claim floating alone.

Trust test: For prompts containing “licensed / 持牌 / registered,” does your public story include a stable registry identifier, not only star ratings?

See [Official registry vs Google Business Profile](/blog/official-registry-vs-google-business-profile-for-ai).

Layer 4 — MCP and Public APIs: How Agents Call You

Model Context Protocol (MCP) tools let agent runtimes query Trusgent without scraping HTML:

•Search / get agent profiles

•Verify trust signals

•Read business facade payloads for booking context

You do not need to host MCP yourself to benefit: once your Agent Card and Brand Page are complete, third-party agents that already connect to Trusgent MCP can retrieve you as a structured object.

AAO test: Is there a documented path from “shortlist item” to “action payload” (hours, contact, booking URL)?

Where JSON-LD Fits (Your Own Site)

Keep publishing schema on your corporate site:

•`LocalBusiness` / `Restaurant` with identical hours

•FAQPage for AEO snippets

•Consistent `@id` and sameAs links to your Trusgent Passport and registry page when appropriate

JSON-LD helps search engines extract; Agent Cards help agent runtimes select. Do both when budget allows; if you must prioritize one new surface this month, finish `/t` + `/b` first.

14-Day Implementation Plan

Days 1–3: Claim `/r` (if applicable); unify legal/trade name.

Days 4–7: Complete Agent Card services + regions; enable Directory readiness.

Days 8–10: Publish Brand Page hours/menu/contact; sync with GBP.

Days 11–14: Start authenticity verification; run the [weekly AI visibility prompt protocol](/blog/ai-visibility-weekly-prompt-protocol-for-merchants) to measure mentions.

Common Failure Modes

| Failure | Symptom in AI answers | Fix |

| --- | --- | --- |

| Empty capabilities | “I found a listing but no details” | Add services / products |

| Hours conflict | Hedged or wrong open/closed | Diff GBP vs `/b` |

| No provenance | Skipped on “licensed” prompts | Claim `/r` + verification |

| Pretty page, no JSON | Cited once, never invoked | Complete Agent Card + facade |

| Fake urgency CTAs | Agents refuse to recommend | Prefer honest booking paths |

FAQ

Do I need engineers to use MCP?

Not to be callable. Completing your Trusgent profile exposes APIs/MCP-friendly data. Building a custom MCP server is optional and usually unnecessary for merchants.

Is Agent Card the same as schema.org JSON-LD?

Related goals, different primary consumers. JSON-LD targets crawl/extract ecosystems; Agent Cards target agent directories and tool calls. Overlap in facts is good; duplication of conflicting facts is bad.

What should restaurants prioritize first?

Identical hours everywhere, one clear booking/WhatsApp path on `/b`, and a claimed FEHD (or relevant) registry link on `/r`.

Closing

Treat Agent Card + Brand Page + registry provenance as your minimum viable Trust Layer. JSON-LD on your site and MCP connectivity amplify it. Merchants who finish this stack give assistants fewer reasons to invent—and more reasons to shortlist and call them. For the visibility ladder, revisit [SEO vs AEO vs GEO vs AAO](/blog/seo-aeo-geo-aao-practical-matrix-for-local-merchants).