← Work

Agent tooling · live commerce

ListingRizz

MCP listing agent plus a real product shell — then put into production against Stripe. Agents get tools. Operators get auth, billing, vault state, and measurable sell-through.

Problem

Marketplace listing work is slow, inconsistent, and usually stuck in chat demos or spreadsheets. Agent tooling collapses the moment it needs auth, money, durable vault state, and a path from draft listing → live inventory → cash.

What was built

Express MCP server on Supabase + Zod with vault schema tooling (check/migrate scripts), paired with a Next.js operator site using Clerk, Stripe, Vercel Blob, React Email, and mcp-handler — so listing agents write into a real product path instead of a local toy. Operated that loop into live Stripe sell-through: listing generation, vault-backed inventory state, and checkout volume tracked in Stripe Dashboard.

Who used it

Operators running listing workflows that need agent tools with real auth and billing — and a storefront that has to clear cash, not just generate copy.

Measurable results

Stripe net volume
CA$12,856.31
Window
Aug 12 – Aug 25, 2026
Prior period (same length)
CA$0.00
Peak day
CA$3,126.43 (Aug 23)
Daily average
~CA$918
Agent surface
listingrizz-mcp + vault scripts
Product shell
Clerk · Stripe · React Email

Stripe sell-through

Stripe Dashboard net volume chart for ListingRizz, August 12–25 2026, totaling CA$12,856.31 versus CA$0 previous period
Stripe Dashboard · Net volume · Aug 12–25, 2026 · CA$12,856.31 · prior period CA$0.00 · peak CA$3,126.43 on Aug 23

Net volume from Stripe Dashboard (Aug 12–25, 2026 daily series). Prior comparable window was CA$0.00 — this is the first measured sell-through period after the agent + product shell went live against the store.

The constraint

An MCP listing agent with no product shell is a local script. A product shell with no vault schema is a chat UI. Neither clears cash until listing state, auth, billing, and the storefront share one operable loop.

The decision

Ship the MCP server and vault tooling first, wire Clerk/Stripe/email into the Next.js surface the same week, then run the agent against live inventory instead of waiting for a perfect catalog — measure sell-through in Stripe, not prompt screenshots.

Architecture evidence

  • Stripe Dashboard: CA$12,856.31 net volume over 14 days (Aug 12–25, 2026); prior period CA$0.00; peak CA$3,126.43 on Aug 23
  • listingrizz-mcp server with vault check/schema scripts on Supabase + Zod
  • Next.js product site: Clerk auth, Stripe money path, Vercel Blob, React Email, mcp-handler
  • Operated listing → vault → storefront loop into measured commerce, not a demo transcript

What I'd do differently

I would publish the MCP tool list as a versioned contract the same week the vault schema landed, and instrument listing→order attribution earlier so every Stripe dollar maps to a specific agent run.

MCPExpressNext.jsClerkStripeSupabase

Related dossier entries: ListingRizz.

Back to work