Clover MCP
Server
I built an open-source Python server that lets AI assistants answer questions about Clover point-of-sale data and make selected changes with operator controls.
“Which items are low on stock?”
- AI clientCalls
list_low_stock_items - Clover MCPReads Clover data with the merchant’s credentials
- ResultReturns a structured list to the client
- Role
- Author and maintainer
- Stack
- Python, FastMCP, Clover REST API
- Delivery
- MIT-licensed Python package
From merchant questions to reliable answers
A merchant might ask which items are running low or how refunds affected daily sales. The server exposes focused tools through the Model Context Protocol (MCP). It handles API calls and calculations, then gives the assistant structured results it can explain.
The browser demo uses a scripted assistant, a Clover sandbox catalog snapshot, and synthetic sales and stock data. Price changes are simulated in the demo page.
Engineering decisions
Calculate sales in code
Clover exposes refunds separately from payments. The sales tool totals successful payments, subtracts refunds, and reports voids separately. The assistant can then explain the result in the merchant’s currency.
Sales calculation notesGuard changes close to the API
Clover treats a stock update as an absolute quantity. I added a dry run and a pre-check against a fresh inventory read. A change can still occur between the read and the write. Operators can disable writes entirely.
Inventory API auditSupport local and hosted clients
A local client can run the server over standard input and output for one merchant. Self-hosted HTTP adds OAuth token validation and can map authenticated clients to their merchants. Operators take on identity and token management in exchange.
Deployment optionsWhat shipped
I published the server on PyPI with the source and test suite on GitHub. The sandbox audit records observed response shapes and identifies endpoints that returned empty results during testing. This is an independent open-source project for the Clover API.