Imagine opening a regular chat — the same ChatGPT or Claude you ask for recipes and email drafts — and typing one sentence: “Create an invoice: this customer, HP printer, 10 units — and show me the current exchange rates in the system.” Half a minute later a ready invoice sits in your accounting, and the chat shows a rate table.

This is not a concept or a “someday” promise. These are screenshots of ERPJS actually working inside the ChatGPT and Claude desktop apps — the ones anyone can install on any computer.

What happened

ERPJS now connects to ordinary AI chats. Not to special “agent systems”, not to developer tools — to the apps half of business owners already have on their computers.

Technically it works through an open interface (called MCP) — the same one from our case study about a store owner who linked his website to his accounting by himself. But where he built an integration, here there is nothing to build: connect once and work.

The point: the entry barrier is gone

Our previous AI case studies had a hidden barrier. The agent created documents — but through a separate tool. A client built an integration — but the agent had been set up for him. The reader thought: “Cool, but that’s not for me.”

Now there is no barrier:

• In Claude, connecting your own system is available even on the free plan — one custom connection is allowed, and one is exactly what a business needs: its accounting.
• In ChatGPT — it works too: we will guide you through the details for your plan when you connect.
• No developer needed. Setup is two fields in a dialog: a name and an address. Once — and forever.

What working with it looks like

Like ordinary messaging. You write in plain language — the system acts:

• “Create an invoice for 10 printers for customer N” — the invoice is created, with a number, date and total;
• “What are the latest exchange rates in the system?” — a table: USD, EUR, GBP;
• “How much stock is left in the warehouse?” — actual balances from the records, not from memory.

One detail we particularly liked (you can spot it in the screenshot): when the assistant was asked for exchange rates, it noticed by itself that the records in the system were not from today and suggested updating them. It is not a blind executor — it sees the state of your records and points out what deserves attention.

As much control as you want

Documents from the chat arrive in the system as drafts — with an “unposted” status. A handy option while you are getting started: review every document before posting until the assistant earns your trust, then simply post right away. AI prepares the documents; how much control you keep is up to you.

The connection works only through your access to your own system: the chat sees exactly what your ERPJS account is allowed to see.

How to connect

In Claude (desktop app or claude.ai) — three steps:

1. Settings → Connectors → Add custom connector;
2. enter a name (say, “My ERPJS books”) and the connection address;
3. click Add. Done — new chats can see your accounting.

In ChatGPT — slightly longer: Settings → Apps → Advanced settings → Create app → a name, the server address and the authentication method. The new app then shows up in your Apps list — and ERPJS is available in your chats.

The connection address for your company is provided by ERPJS — contact us and we will send it together with a step-by-step guide.

Why it matters

We have been opening ERPJS to AI step by step: first an agent analysed data, then it created documents, then a client built his own website integration. Now the next step: accounting arrives in the app you already use every day.

For a small business it means one simple thing: between “I need to issue an invoice” and a ready invoice there is a single chat message. No new software to learn. No developer. No automation budget.

FAQ