What breaks first at ten client files is not capability, it is ambiguity. The assistant can hold as many connected companies as you connect. What it cannot do is guess which one you meant when two of your clients are numbered companies with nearly identical names. Almost everything below is about removing that guess.
Name the company, every time
If more than one company is connected, the company has to be named in the request. Not implied by context, not carried over from the previous question, named.
Three habits make that reliable in a practice:
- Start each session by asking what is connected and at what access level. You get the list of company names as they are actually registered, which is the vocabulary the assistant understands.
- Agree on one shorthand per client and use only that. If two entities share a legal name root, decide which words distinguish them and always say those words. Half a name is where the wrong file gets opened.
- Check the first answer names the company back. If it does not, ask which company it used before you read any numbers.
Then add the verification habit covered in the guides index: ask something you already know the answer to, on the file you think you are in. It catches a wrong company in seconds.
Access levels, per file
Access is per company, so a practice can hold most files read-only and open exactly one for the work in front of it. That is the pattern worth defaulting to: nine client files read-only, the tenth in full access for as long as the task lasts, then back.
Every connection starts read-only, so this is where you begin whether you plan it or not. Sending a document by email, voiding and deleting each need an explicit confirmation in the conversation, even on a company with full access.
The weekly sweep
This is the part that pays for itself in a multi-file practice. One session, once a week, the same questions asked of each file in turn:
- What changed since last Friday.
- Aged receivables, and anything past the threshold you care about.
- Aged payables, and what is due in the coming two weeks.
- Bank and credit card balances against what you expect.
- Anything sitting in the accounts you use as a holding place.
Paste the answers into your working notes as you go. Nothing here is clever. The reason it saves real time is arithmetic: assembling those four or five reports by hand is a short sequence of clicks, and you were doing it once per file. The assistant does not make any single lookup dramatically faster. It makes the twelfth one cost the same as the first.
Two things to keep it honest. Ask the same questions in the same order each week, so a missing answer is obvious. And treat anything the sweep flags as a pointer to a file you should open, not as a conclusion.
Where it does not help
A messy file produces confident answers built on the mess. If a client's ledger has a year of miscoded transactions, faster access to those transactions is not progress, and the assistant will not tell you the underlying data is wrong. Fix the file first.
If your bottleneck is data entry rather than retrieval, the gain is small. This is strongest on questions, comparisons and drafts, weakest on volume keying.
And month-end judgment does not compress. Accruals, cut-off, and anything that depends on what you know about the business still take the time they take.
What it costs at practice scale
39 $ CA per company per month, and 29 $ CA per company from ten companies. Billed monthly through Stripe per connected company, cancel anytime with access to the end of the paid period. There is a 14-day free trial with no credit card, which is enough time to run two weekly sweeps and see whether the habit sticks.
The honest warning about per-company pricing: connect the files you actually work in. A dormant file you open twice a year does not earn a monthly line, and connecting your whole client list because it is easy is how the bill stops making sense. Connect five, run them for a month, and add more only when a file has earned it. The pricing page has the current terms.
Where this is the wrong choice
If most of your files are QuickBooks Desktop, this does not apply. Desktop does not expose the API this depends on, so it is not supported and cannot be.
If your files are small and you touch each one a few times a year, per-company pricing works against you and the sweep has nothing to sweep. A practice with a handful of active files and a long tail of dormant ones should connect only the active ones, or skip it.
If you are technical, comfortable running a process on your own machine, and only ever work in one company file, Intuit's open-source MCP server is free and covers create, read, update and delete across 29 record types and 11 reports. It needs a developer account, an OAuth app, API keys and a process kept running, and it does not work from Claude on the web or on a phone. For one file, that may still be the right trade. The comparison works through it, and the setup guide covers connecting the first company if you decide to go ahead.