Hey all — John here, first post. I’ve been using ERPNext since v14 and currently run ERPNext v16 — Just a one-person shop with a few small entities under it. Real books, Real Problems, and Real Trust issues when it comes to AI and my wife’s money.
I saw the MCP for ERPNext thread from earlier this year, so I know I’m not the only one here thinking about letting agents talk to ERPNext. That’s exactly the rabbit hole that produced the thing I want to share.
The more automation being pointed at ledgers, the less comfortable people get with “trust me, it posted the right entry.” I wanted proof, not trust. So I built Pacioli — a governance layer that sits between anything (an MCP client, an A2A agent, a plain API call — any door) and your ERPNext books:
- Immutable ledger — nothing quietly rewrites history.
- Plan → prove → record — every change is planned, checked against the books, and leaves a receipt you can verify after the fact.
- Any door — same governance whether the caller is an AI agent over MCP or a script over REST.
Named after Luca Pacioli, the friar who wrote the book on double-entry in 1494 — felt right for something whose whole job is keeping the ledger honest.
I built it for my own peace of mind, so it’s not a demo. It’s open: GitHub - john-broadway/pacioli: Least-privilege governance for ERPNext — the credential floor + a governed agent front door, MCP · A2A, one spine. No debit without a credit: PLAN · CONSENT · PROVE · UNDO. · GitHub (pip install pacioli pacioli-guard).
Happy to answer anything — and genuinely curious how others here are handling agent access to ERPNext.