Built a 50-second demo video showing how Japan's night-leisure industry SaaS "tasteck" uses ChatGPT MCP to replace a tedious 11-step booking transcription with a single sentence from the owner.
The problem
LINE/email bookings arrive in free-form text. Staff manually copy them into the order screen — date, time, course, cast nominee, customer search, new guest creation, payment, save. 3-5 minutes per booking, 20-30 bookings during peak.
The fix (MCP path)
Owner pastes the LINE message into ChatGPT (or Claude Desktop) and says "put this into tasteck". The MCP tool chain:
- Parses date / time / course / nomination / payment from free text
- Looks up the customer by phone, or auto-creates a guest
- Returns a
dry_runpreview for the owner to visually confirm - Owner confirms → committed to tasteck DB
Safety design (the unusual part)
- Owner's hand LLM only, not staff multi-tenant
- OAuth scope per business owner
- 2-stage commit (
dry_runpreview → confirm) - No booking data is retained by the external LLM vendor
Watch the demo
More
tasteck.tech — vertical B2B SaaS for Japan's night-leisure industry (massage / hostess club / host club).
This is a build-in-public log; happy to answer questions in the comments.
Top comments (0)