← Work
FactureCraft
Local quotes and invoices, for people and for their tools.
- Design and development
- Personal open-source project.
- Active
The problem
Online invoicing tools keep the data on their side and are hard to drive from other tools. The goal: prepare a quote from an AI assistant while keeping control over what actually goes out to a client.
Constraints
- Data stays on the machine (local SQLite).
- An assistant can prepare a draft, but only a person can finalise: neither the CLI nor the MCP server can issue a document.
- Legal numbering must continue the existing series imported from Facture.net.
- Factur-X compliant documents.
The solution
A Cargo workspace with three interfaces on the same business core:
| Interface | Role |
|---|---|
| Native app | Dashboard, documents with live preview, catalogue, finalisation, payments |
facturecraft CLI |
Prepare drafts, derive documents, PDF previews, import, backup |
facturecraft mcp |
Search contacts and products, read documents, prepare drafts |
What was built
- Contacts, quote lines with taxes and discounts, PDF previews.
- Import of Facture.net exports with preview and confirmation.
- Quote-to-invoice conversion, deposits, payments, credit notes.
- Real numbering that continues the imported series, with a dated PDF frozen at finalisation.
Results
The first five roadmap stages are in place. Historical reconciliation and credit notes are covered by synthetic regression tests, and Factur-X validation is documented in the repository. macOS packages are not yet signed; Windows and Linux remain to be checked.
What it shows
How to open a business tool to AI agents without handing them the keys: one source of rules, three interfaces, and a clear line between preparing and committing.