@timothy sorry I may have given you false hope using āintegrationā word but Iām happy to help
I originally had an import script. And I just finally made a custom integration (more below). The script was JS on Google Apps Scripts which runs a daily import to Fibery. It is only importing invoices (ACCREC) (with only limited data about amount, status, etc) and some related Contact information.
I then worked on a proper Custom Integration and only just now deployed it working. I used the exact method from the Fibery team AI Custom Integrations. I commented here about how well it works on my first attempt and it still applies
š¦ Custom Fibery Integration Development with AI - #11 by mdubakov.
However the integration is the exact same limited scope as the script.
But one thing I did was I spent sooooooo long on getting the OAuth to work. In fact that was the only real bug. Iām sure a real developer could have isolated the issue faster and more programatically tested. But it seems to be some quirks with how Xero works with multi-tenants and some comments in the Fibery documentation, and I have no idea what Iām doing ![]()
Below are some rambling tips. I sure Iām telling you things you already know generally about software dev as well:
- I used Claude for everything. You may not need it but Opus 4.5 because I have Max plan so why not.
- Plan phase output to an md file so I could read more deeply and make specific manual edits to the plan
- Itās really limited in scope compared to the breadth of the Xero API docs so that helped to keep it focused. Only 2 DBs with Invoices and Contact both with limited fields.
- I donāt know how but you should have very verbose logging on the OAuth to include all steps in the chain. I spent a lot of time with piecemeal changes and re-test, fail, analyse the logs.
- The new MCP tool for Claude will help to read the Fibery user guide. I forgot it was released last week and it may have saved some time. I could have worked around to get the help docs in but to be honest I just forgot.
- I didnāt realise until later that the free tier Vercel only streams logs (canāt query after) for up to 5 mins so I was wasting time in this loop. Basically, if you can give LLM direct access to query the logs then the debug loop will be much shorter. Or just pay and donāt be cheap

I hope that helps! And Iām happy to share more if I missed something.