How to connect QuickBooks Desktop to ChatGPT or Claude

Why there is no direct connection

Claude and ChatGPT reach outside data through connectors. Intuit built connectors for Online because Online lives on Intuit’s servers. Desktop lives on your machine, and its only doors out are an old developer toolkit that runs on the same Windows computer and the reports it exports.

Developer products exist on that toolkit. They are for programmers, and they need software installed next to QuickBooks.

The route that works

1. Export three reports: the customer list, the item list, and a sales-by-customer detail report with the dates typed in.
2. Load them into a database. Every month’s total is checked against QuickBooks’ own summary before anything publishes.
3. Add the database’s connector to Claude or ChatGPT, sign in, and ask. It can look and cannot touch. There is no path back into QuickBooks.
4. Repeat the three exports every week.

It works in Claude, including the Claude app on your phone. It works in ChatGPT on the web if you are on a ChatGPT Business plan, or through a GPT we publish for you; ChatGPT on the phone is not something we promise today.

What it is like

You type the question the way you would say it out loud. The answer comes back in plain English with a table. Ask how it got there and it shows the computation. Here are two questions in a row, asked of a fictional company. The second is a follow-up.

with last_order as (
    select entity_id, max(date) as last_date
    from txns where txn_type = 'Invoice'
    group by entity_id),
year_before as (
    select t.entity_id, sum(t.total) as rev
    from txns t
    join last_order lo on lo.entity_id = t.entity_id
    where t.date > (cast(substr(lo.last_date, 1, 4) as int) - 1) || substr(lo.last_date, 5)
    group by t.entity_id)
select e.canonical_name as customer, e.customer_type, e.phone, e.email,
       lo.last_date as last_order, round(yb.rev) as their_last_12_months
from entities e
join last_order lo on lo.entity_id = e.id
join year_before yb on yb.entity_id = e.id
where lo.last_date < '2025-07-27' and yb.rev > 5000
order by their_last_12_months desc

with per_year as (
    select year, customer, sum(amount) as rev
    from v_sales_lines
    group by year, customer),
totals as (
    select year, sum(rev) as total from per_year group by year),
ranked as (
    select year, customer, rev,
           row_number() over (partition by year order by rev desc) as rk
    from per_year)
select r.year, round(t.total) as total_sales, round(sum(r.rev)) as top_10_sales,
       round(100.0 * sum(r.rev) / t.total, 1) as top_10_share_pct
from ranked r
join totals t on t.year = r.year
where r.rk <= 10
group by r.year, t.total
order by r.year

Northgale Building Products is a fictional company. Every name, figure and supplier on this page is made up, built to behave like a real three-and-a-half-year QuickBooks Desktop file so we can show real screens without showing anyone’s real numbers. Data as of the 2026-07-27 refresh; the last invoice in the file is dated 2026-07-24.

Short answers

• Does Intuit’s ChatGPT connector work with QuickBooks Desktop?No. It is built on QuickBooks Online. Desktop and Enterprise are not covered.
• Can ChatGPT read my QuickBooks company file if I upload it?No. Only QuickBooks opens a .QBW file. Exported reports are what an assistant can read.
• Can the AI change anything in my books?No. The database is built from reports QuickBooks printed. There is no connection back.

Related questions

• Is there an MCP server for QuickBooks Desktop?
• Does QuickBooks Desktop have AI?
• A live dashboard from QuickBooks Desktop, without a connector subscription
• Your systems, joined