MCP overview
MCP is the agent access layer. The sellable product is not the protocol itself; it is the private memory, corpus, fuzzy matching, ontology, and recommendation logic exposed through safe tools.
Audiences
| Audience | Example tools | Commercial path |
|---|---|---|
| Individual user | get_collection, suggest_next_wear, log_wear | Power or Connoisseur |
| Developer | search_fragrance, match_fragrance, get_similar | Developer MCP key |
| Retailer | gift_recommendations, map_sku_to_fragrance | Widget or Shopify plan |
| Enterprise agent | Scoped corpus and trend tools | Contracted MCP access |
| Sniffopotobot | Internal and user-facing oracle tools | Product intelligence |
Principles
- Read access is the default.
- Write scopes are explicit opt-in.
- Partner and couple data needs consent.
- Every agent action must be auditable.
- Tool results must include confidence and source tier where fragrance facts are returned.
- Commercial keys must have usage metering and rate limits.
Planned commercial packages
- Consumer MCP read access bundled into Power/Connoisseur.
- Developer MCP subscription for external builders.
- Retailer MCP for embedded AI shopping agents.
- Enterprise MCP for brands, marketplaces, and data licensing.
- Future Shopify Sidekick-like integration once the Shopify app is mature.
Setup guides
Start with Connect AI clients, then use the guide for your client: