Case studyHoldings · Product design

MCP invoicing

LaunchedSeptember 2026

BuiltAn MCP server

Sign-up designed inVerbal

Works withClaude · ChatGPT · Custom agents

Connect your AI agent · the first step of setup

Problem

Once we entered invoicing, customers had the same two complaints: creating invoices still took too long, and following up on them was frustrating.

5 customers interviewed

We spoke with five customers who invoice often. Sending, resending and tracking invoices was tedious, and nobody enjoyed it.

WHAT WE HEARD

Everyone had a story of leaving money on the table because they couldn’t track an invoice down.

WHAT WE HEARD

Sending, resending and following up was the part people wanted gone, not the invoice itself.

When we launched, there were no other MCP invoicing tools on the market.

LOOKED ATOther MCP toolsHow they explained what an agent could do.

LOOKED ATSign-up flowsHow other products got a connection set up.

Customers already using agents will want to use them to send their invoices.

We built an MCP server that connects to any agent: Claude, ChatGPT or a custom one. Customers send invoices with a prompt.

Sending an invoice from a prompt.

END TO END · HOW AN INVOICE GETS SENT

01ConnectGenerate a key in Holdings and paste the connection URL into your agent.
02Prompt“Send Acme the retainer invoice. Follow up with Miller.”
03Agent draftsThe agent calls the MCP server to build the invoice from your customers and items.
04You approveNothing goes out without your yes. Every action is logged.
05Sent and trackedHoldings sends it, tracks it and chases it; paid invoices post to your books.

SETUP FLOW · 3 STEPS, ABOUT 2 MINUTES DESIGNED IN VERBAL

First 60 days

INVOICES CREATED · 60 DAYS96invoices created through agents in the first 60 days

ACTIVE CUSTOMERS3customers using the product several times a week