How we made our niche-industry SaaS MCP-ready (and watched ChatGPT call our dispatch tools)
tasteck, a B2B SaaS serving the Japanese night-leisure industry (dispatch and cast shift management) with 8 years of operational data across ~100 venues, became MCP-ready. Two days after the MCP design post, ChatGPT Plus could invoke live tools such as list_available_drivers, list_cast_shifts, and list_assignable_casts, returning natural-language replies. The team estimated a two-week B2 OAuth sprint (June 16–July 1) but completed it in one day by carefully reading the spec before coding. They encountered 12 distinct traps between OAuth issuance and successful tool invocation, all caught by QA logs. The MCP server uses @modelcontextprotocol/sdk with SSE transport wired into a NestJS controller. Three read tools (B1) were shipped: list_available_drivers, list_cast_shifts, and list_assignable_casts (which joins roster, stage-name set, and shop match). A natural-language date helper, resolveBusinessDate(naturalText, company), handles relative dates like 'today' and per-tenant business-day boundaries (e.g., day flips at 04:00 or 05:00, configured per Company.changeDateTime).
Shows how niche SaaS can quickly integrate with AI agents via MCP, reducing integration time from weeks to days.