rapido
Book a demo
◂ Back to blog
· updated

How to connect Accredo to ChatGPT

Use ChatGPT as a conversational interface to your Accredo company file — query debtors, draft orders, run reports — via Custom GPTs and the Rapido MCP server.

ChatGPT is on most New Zealand business desks. The accounts team uses it for email drafts, the ops team uses it for SQL, the GM uses it for board prep. What if it could read straight from your Accredo company file when you asked it a question?

That’s what the Rapido MCP server gives you. ChatGPT — running through ChatGPT Apps (formerly “connectors”) or the OpenAI Apps SDK — connects to your Accredo over the open Model Context Protocol, and starts answering questions you’d otherwise have to extract manually.

What “ChatGPT connected to Accredo” looks like in practice

A few real prompts customers run today:

  • “What’s our exposure on Mainland Steel right now? Pull AR aging plus open OE drafts.”
  • “Find every product in IC where we’ve got more than 6 months of stock at current run-rate.”
  • “Summarise this month vs last month for the Saturn-side branches.”
  • “Draft a remittance for the matched supplier invoices over the last fortnight.”

ChatGPT writes the natural-language answer; behind the scenes it’s calling Rapido’s MCP tools (/debtors, /products, /orders, /invoices) and getting structured JSON back from your live company file.

What you need

  • A ChatGPT Plus, Team, or Enterprise subscription (Custom GPTs / Apps live there).
  • A Rapido account with the Rapido MCP server enabled.
  • A bearer token, issued by Rapido, scoped to your Accredo company file.

That’s it. No middleware to install, no hand-rolled OData clients, no Zapier chain.

Setting it up

For ChatGPT Apps (the renamed connectors surface inside ChatGPT), add an app in Settings → Connectors / Apps pointing at:

https://mcp.rapido.co.nz/v1

Authenticate with the bearer token we issued. ChatGPT discovers the available tools (debtors, creditors, products, orders, invoices) automatically and exposes them to the model.

For the OpenAI Apps SDK (build an MCP-aware agent on top of the API directly), use the same URL and token in your mcp configuration block. The protocol is the same; only the host application changes.

ChatGPT can look up a customer, check stock, run a query, summarise a period — high-leverage workflows that put data in front of the right people without rebuilding any UI.

Writes (drafting an order in OE, marking a remittance) are explicit. You turn them on per workflow, scoped per token. A token can have read-everywhere + write-only-to-OE-drafts, for example. Audit trail is in Accredo and on our side.

Where the data goes

Customer data passing through ChatGPT is governed by OpenAI’s API and enterprise terms. OpenAI doesn’t train on API or business-tier data by default. The MCP server itself runs on AWS Sydney (ap-southeast-2).

If your business has a hard rule against US-based AI providers seeing customer data, the same MCP server works with Claude (Anthropic) — see How to connect Accredo to Claude. Many customers run both, with different teams using different clients.

What ChatGPT can’t do (yet)

ChatGPT’s MCP support is excellent for read-heavy and conversational workflows. For long-running agentic workflows — “watch this inbox forever and draft orders into Accredo as POs arrive” — you want a purpose-built agent. That’s what our PO2Order product is.

ChatGPT is best for the day-to-day “what’s going on in my ERP?” questions where you want a fast natural-language answer.

Get started

Book a 30-minute discovery call. We’ll connect ChatGPT to your Accredo, run a few real queries against your data, and you can decide if it’s worth a token. No services engagement, .

Further reading