Month-End Close with Claude Cowork: A Step-by-Step Guide

Month-end close is the part most people dread. You match your books against your payment processor, then chase mismatches and missing receipts by hand. Claude Cowork has a skill for exactly that, /close-month, and this guide walks the whole run, one level deeper than the close section in the full Claude Cowork for accounting and bookkeeping guide.

What is a Claude Cowork month-end close, and how does /close-month work?

A Claude Cowork month-end close is a reconciliation run inside Claude's Small Business plugin. Anthropic's official line: /close-month "reconciles your books against your payment processor and writes the close packet for your accountant." You need the Claude desktop app on a Pro, Max, or Team plan, with QuickBooks connected. It matches your QuickBooks transactions to processor settlements by amount and date. Then it flags four things: mismatches, uncategorized transactions, duplicates, and missing receipts. You clear each flag in plain English. It hands back two files, a reconciliation spreadsheet and a one-page close packet. It is a checklist helper, not an auto-close.

Key Takeaways

• You need three things: the Small Business plugin, a Pro, Max, or Team plan on the desktop app, and QuickBooks connected. Add PayPal if you run payments through it.
• Two ways to start it: the /close-month slash command, or plain English like "close out this month and tell me what needs my attention."
• It flags four problem types: mismatches, uncategorized transactions, duplicates, and missing receipts. You clear each one yourself.
• You get two outputs: a reconciliation spreadsheet and a one-page close packet for your accountant.
• It reconciles, it doesn't accrue. Accruals, prepaids, and depreciation stay your judgment calls, not part of the run.
• Nothing posts or sends without you. Claude drafts and flags. You approve.
• Use a Project to make fixes stick. One per recurring client keeps last month's corrections loaded for the next close.
• You can schedule it monthly with /schedule, and it runs with no device online.

What /close-month actually does

Claude Cowork bundles a set of money skills called the Small Business plugin. /close-month is the one built for your monthly close. The official description is short. It "reconciles your books against your payment processor and writes the close packet for your accountant."

Under the hood, it runs a routine called month-end-prep. It reads your QuickBooks data, plus PayPal if that is one of your processors. That is the raw material for the match.

Here is the honest framing. /close-month does the tedious matching and the first-pass flagging. It does not close your books on its own. You stay in the loop for every judgment call. Think of it as a sharp checklist that fills itself in, not a button that closes the month while you look away. If you are new to the plugin, start with how to use Claude for bookkeeping for the wider tour, then come back here for the close.

Before you run it: what you need

A few things have to be in place first. None of them take long.

You need the Claude desktop app, on a Pro, Max, or Team plan. The Small Business plugin lives on those plans only. To install it, open Cowork, go to Customize, then Plugins. Click the plus, pick Small Business, and hit Install.

You need QuickBooks connected. One caveat worth saying plainly, not burying in a footnote: the QuickBooks connector is US customers only right now. You do not need an active QuickBooks subscription to connect it. You can supply your business type, your region, and about ten transactions instead, and Claude has enough to work with. For the full connect walkthrough, see how to connect QuickBooks Online to Claude.

PayPal helps too, if you take payments there. /close-month reads QuickBooks and PayPal together. That is how it matches your books against what actually settled.

Run /month-heads-up before you close

There is a companion skill worth running first. It is called /month-heads-up.

Where /close-month looks back at the month you just finished, /month-heads-up looks forward. It reads the next 30 days of cash, finds your tightest week, and flags what to watch before month-end. It runs on your QuickBooks data.

Why pair them? A close tells you what happened. A heads-up tells you what is coming. Run the heads-up first, and you walk into the close already knowing where cash is tight. You are not surprised by a slow week after the books are done.

Prompt: Give me a heads-up on the next 30 days of cash and point me at the tightest week.

It takes a minute. Then you close. The two skills work as a pair. Look forward, then close the month behind you.

How to run a Claude month-end close

You have two ways in. Both land in the same place.

The first is the slash command. Type /close-month and point it at the period you just finished. The second is plain English. You do not have to memorize the command name.

Prompt: Close out this month and tell me what needs my attention.

Either way, Claude reads your QuickBooks and PayPal data, runs the match, and comes back with a list of flags. A Claude month-end close comes down to that one step. Run it, then clear what it surfaces. You are not typing out each instruction. You start it, and it works through the reconciliation on its own, one flag at a time.

What it checks under the hood

The match itself is simple to describe. /close-month lines up each QuickBooks transaction against your processor settlements. It matches on two things: amount and date.

When something does not line up, it gets flagged. There are four kinds of flag.

A mismatch is a transaction that does not tie to a settlement, or a settlement with no transaction behind it. An uncategorized transaction is one with no account assigned yet. A duplicate is the same charge posted twice. A missing receipt is a charge with no document attached.

That is the whole check. Amount and date go in. Four flag types come out. Nothing gets changed at this stage. You just get a clean list of what needs a decision.

A worked close: clearing the flags

Here is what a flag list looks like on a made-up month. The numbers below are illustrative. Yours will be different.

Say /close-month turns up four items.

• A PayPal settlement has not matched any QuickBooks deposit yet. Flagged as a mismatch.
• A $340 charge posted twice. A plain duplicate.
• A $58 charge from a new vendor. Uncategorized.
• A $210 hotel charge with no receipt attached.

You clear each one the way you would explain it to a coworker.

Your replies: "The PayPal settlement is still in transit, give it a day." "The duplicate is a sync glitch, delete the second one." "The $58 charge is software, code it to subscriptions." "The hotel was a client trip, I will upload the receipt tonight."

Claude updates the reconciliation spreadsheet as each flag clears. When the list is empty, it builds the close packet. The dollar amounts here are made up, and your real flags carry your own numbers. The pattern holds no matter how many a month turns up. You clear each one in plain English. You are not hunting through a spreadsheet row by row.

The difficult 20%: where a close actually gets hard

Most of a close is easy. The match ties, the category is obvious, you move on. The hard part is a small slice, maybe a fifth of the work, that eats most of your time. Growthy calls it the difficult 20%. Here is how it shows up in a close, and what you decide each time.

Net deposits versus gross sales. Your payment processor takes a fee before the money lands. Say you sold $1,000. PayPal keeps its cut, and $971 hits your bank. Your books may show the full $1,000 in sales. That $29 gap is the fee. /close-month sees the deposit not matching the sale and flags it. You decide the fix. Record the gross sale, then book the fee as a processing expense. The deposit ties once you split it. This is gross-versus-net, and it is the single most common close mismatch. The same logic holds for any processor. /close-month reads QuickBooks and PayPal directly. For a processor it does not read, you bring the fee numbers in yourself, but the decision is identical.

Transfers that look like income. You move $2,000 from checking to savings. Your bank shows money leaving one account and landing in another. Neither line is income. Neither is an expense. It is your own money changing seats. /close-month may flag both sides as uncategorized or as a mismatch. You code it as a transfer, not revenue and not spending. Miss this, and your income looks $2,000 bigger than it is. This is the flag that most often hides a real error behind a clean-looking total.

The blank ACH line. A charge posts as "ACH PAYMENT" with no vendor, no memo, nothing. The bank feed dropped the detail. /close-month flags it uncategorized because there is no way to guess the account. You recognize it. That is your insurance carrier. You supply the category. Claude cannot know what the bank did not send. You can.

None of these are Claude failing. They are the judgment part of a close, the part that was always going to need a person. /close-month finds them and lines them up. You make the call.

What a full close needs that /close-month skips

Reconciliation is the biggest chunk of a close. It is not the whole job. /close-month does the reconcile-and-flag slice well. It does not do the rest, and it is worth being clear about what the rest is.

A complete monthly close usually needs a few adjusting entries that no reconciliation will surface:

• Accruals. You owe for something you have not been billed for yet. The work happened this month, so the expense belongs this month. That is a judgment call and a journal entry. /close-month will not make it.
• Prepaids. You paid a year of insurance up front. Only one month's worth belongs this month. The rest sits on the balance sheet and releases over time. That is your entry to record.
• Depreciation. You bought equipment. Its cost spreads across the months it is used, not all at once. /close-month does not track a depreciation schedule.

These are your entries to make. /close-month reconciles the cash and the categories. It does not decide that half of a prepaid should move, or that this month owes for unbilled work. Those are accounting judgments.

Here is the honest workflow. Run /close-month to reconcile and clear the flags. Then make your period-end adjustments on top: accruals, prepaids, depreciation, any reclass you know is coming. The skill handles the mechanical match. You handle the judgment layer. If you are not sure which adjustments your books need, that is a question for your accountant, and the close packet gives them exactly what they need to answer it.

What you get: two outputs

Every clean /close-month run hands you two files. They map to the two jobs a close actually has.

The first is a reconciliation spreadsheet. It shows every match, and every flag you cleared, with your reasoning attached. That is your record for the month. If a number gets questioned later, the answer is right there.

The second is a one-page close packet for your accountant. It is the short summary they actually want. What closed, what moved, what needs a note. No more forwarding a messy export and hoping they piece it together on their end.

Here is what the close packet pulls together. The period it covers. The reconciliation result, tied out. The flags you cleared, with the reason you gave for each. Anything still open that needs a decision. And a short note for anything unusual, like a fee split or a transfer you recoded. It answers one question fast. Is this month done, and can I trust it? Your accountant does not have to dig through raw exports to find out.

The reconciliation spreadsheet is the long-form backup. The close packet is the summary on top. Together they mean your accountant reads one page first, then drills into the spreadsheet only if a number needs a second look.

Keep your close conventions in a Project

Run a close in a bare session, and it starts from nothing next month. The fixes you made, the way you code a certain vendor, the transfer accounts Claude should never treat as income, all of it resets. You would re-explain the same things every close.

A Project fixes that. A Project holds files, standing instructions, and memory across sessions. Set one up for each recurring client or entity. Put your conventions in it once.

Project instructions: This is Acme LLC. Their chart of accounts is attached. Payments from PayPal land net of fees, so split gross sales from processing costs. Transfers between the operating and reserve accounts are never income. The monthly insurance draft is prepaid, so code it to prepaid expense.

Now every close for Acme starts with that context loaded. Claude codes the recurring vendors the way you taught it last month. It knows which transfers to leave alone. Your corrections stick instead of evaporating.

One Project per client keeps them separate. Acme's rules do not bleed into another client's books. If you close the same handful of entities every month, this is the difference between a close that gets faster over time and one that stays exactly as slow as the first one.

Nothing posts or sends without you

This is the guardrail that matters most for a close. You approve before anything sends, posts, or pays. That holds across every skill in the plugin, not just this one.

Claude reads only what your account can already read. Your existing permissions still apply. Connect a read-only view of a tool, and it stays read-only. Nothing about a close changes that.

Anthropic does not train Claude on your business data. Intuit says your QuickBooks data is not used to train Claude either. Those are specific claims you can check at the source, not vague reassurance. Confirm them yourself before you connect a client's books. For a deeper look at where your numbers go, see how AI bookkeeping tools handle your data.

Schedule it to run every month

You do not have to be at your desk for a close to run. You can put it on a schedule.

Run /close-month once, then type /schedule inside that same task. Or open the Scheduled panel in the left sidebar. Set it to repeat, and Claude runs the close on its own each period.

Prompt: Run /close-month for last month on the 5th of every month, and leave the close packet ready for me to review.

Scheduled tasks run with no device online. Your laptop can be closed. The run still happens on Anthropic's side, and the result waits for you when you check in.

One practical tip on timing. Do not schedule the close for the 1st. Bank feeds and processor settlements take a few days to fully post. Set the run for the 3rd or the 5th, so the month's last transactions have landed before Claude reconciles. A close run too early just flags things that were only in transit.

A full recurring setup looks like this. Run /close-month once by hand, so you have seen it work. Schedule it for the 5th of each month. Point it at the prior month. Tell it to leave the packet ready, not to assume anything. Then each month, you open Claude, find a close waiting with its flags surfaced, and spend your time only on the judgment calls.

Here is the part that keeps it safe. Anything that needs your sign-off does not fire on its own. If a scheduled close turns up flags, they wait for you the next time you open Claude. Nothing posts because a schedule ran while you slept.

A manual close versus a Cowork-assisted one

It helps to see the two side by side.

A manual QuickBooks close goes like this. You open the reconcile screen. You match transactions to the statement one row at a time. You spot the ones that do not tie and dig into why. You hunt for missing receipts across email and folders. You fix categories by hand. Then you export something for your accountant and hope it is clear. The mechanical matching, row by row, is where most of the time goes. The judgment calls are a smaller slice, but they are buried inside all that clicking.

A Cowork-assisted close moves the clicking off your plate. You run /close-month. It does the matching and hands you a flag list. You clear each flag in plain English, the way you would explain it to a coworker. It writes the packet for you. What is left on you is the judgment. Which category is right, whether a transfer is really a transfer, which receipt still needs chasing. That part does not disappear, and it should not.

So the honest contrast is this. Cowork does not remove the judgment. It removes the row-by-row matching around the judgment. Some people report saving real time this way. Your own mileage depends on how messy your books are and how many flags a month turns up. What is certain is where the effort moves. Away from mechanical matching, toward the decisions only you can make.

Tie the close back to your statement

One check outranks the rest. Before you trust the close, tie it to the source.

Pull your bank statement and your processor statement for the month. Look at the ending balances. Then check that your reconciled numbers land on those same figures. If your books say checking ended at $12,430 and the bank statement says $12,430, the month ties. If they are off by $29, you have a fee you did not split, or a transaction still in transit. Chase the difference until it is zero.

This is the moment the whole close earns its trust. A reconciliation that does not tie to the statement is not done, no matter how clean the flag list looks. /close-month gets you most of the way by matching and flagging. This last tie-out is the step that tells you the match is actually right. It takes two minutes, and it catches the errors a tidy total can bury.

What to double-check

/close-month is fast and mostly right. It is not a rubber stamp. A few things deserve your eyes every single time.

Check every category it picks. It may code a charge differently than you would. Say it books a client dinner as "office supplies" because the vendor name was vague. You catch that and recode it. Give transfers between your own accounts a second look too. A miscoded transfer inflates your income without ever looking wrong on the page.

The honest summary: this is a checklist helper, not an auto-close. It does the matching and the flagging. You do the judgment. Treat the output like a sharp junior's first pass. Useful, mostly right, and worth a careful read before you sign off on the month.

Where /close-month stops

/close-month is genuinely good for your own books. It holds up for a couple of simple clients too. This is not a teardown. It does what it was built for, and it does it well.

The limits show up at scale. Cowork works one folder, one session at a time. It does not carry pattern memory per client across a roster. A bare session forgets last month's fixes unless you build a Project to hold them. And there is no built-in audit trail of who approved what, and when, across ten to thirty-plus client books.

That is the exact line where Growthy begins. Growthy connects straight to QuickBooks Online. It uses pattern learning per client, so your fixes stick and compound instead of resetting each month. It reviews every transaction with you before it posts, the same approve-first posture Cowork uses. And it runs ten to thirty-plus client books at once, with an audit trail. Growthy is 85% accurate on first import. You review the rest. And if your work runs into tax season, our sister product TracePrep handles tax workpaper automation.

If you are closing your own books or a couple of simple ones, /close-month may be all you need. If you are running a roster, see how AI bookkeeping tools handle many clients at once before you decide Cowork alone covers it.

Get started with Growthy. Get started and see what it catches on your first close.

Frequently asked questions

These are the questions that come up most about running a Claude Cowork month-end close.

What is a Claude Cowork month-end close?

It is a reconciliation run inside Claude's Small Business plugin. You run /close-month, and Claude matches your QuickBooks transactions to your payment processor settlements by amount and date. It flags mismatches, uncategorized items, duplicates, and missing receipts. You clear each flag, and it writes a reconciliation spreadsheet plus a one-page close packet.

What plan do I need to run /close-month?

You need the Claude desktop app on a Pro, Max, or Team plan. The Small Business plugin, where /close-month lives, is limited to those plans. A Free plan will not show the plugin in Customize.

Do I need an active QuickBooks subscription?

No. You can connect the QuickBooks connector without a paid QuickBooks plan. Supply your business type, your region, and about ten transactions, and Claude has enough to work with. An active plan makes the connection richer, but it is not required.

Does /close-month work outside the US?

Not yet. The QuickBooks connector is US customers only for now. If you are outside the US, the close workflow will not reach your books the way this guide describes. There is no official Xero connector for the money skills either.

What does /close-month actually flag?

Four things. A mismatch is a transaction that does not tie to a settlement. An uncategorized item has no account yet. A duplicate is the same charge posted twice. A missing receipt is a charge with no document attached. It matches on amount and date to find all four.

Does /close-month connect to Stripe or Square?

There is no confirmed Stripe or Square connector for the close. /close-month reads QuickBooks and PayPal. For any other processor, you bring the settlement and fee numbers in yourself. The gross-versus-net logic is the same no matter which processor you use.

How does net-versus-gross work with processor fees?

Your processor takes a fee before the deposit lands, so the deposit is smaller than the sale. Record the gross sale, book the fee as an expense, and the deposit ties. /close-month flags the gap between the sale and the deposit. You decide the split.

Does /close-month handle accruals, prepaids, and depreciation?

No. It reconciles and flags. Accruals, prepaids, and depreciation are judgment entries you record on top of the reconciliation. If you are not sure which your books need, ask your accountant. The close packet gives them what they need to advise you.

What is in the close packet?

A one-page summary for your accountant. It covers the period, the reconciliation result, the flags you cleared with your reasons, anything still open, and a note on anything unusual like a fee split or a recoded transfer. The reconciliation spreadsheet is the long-form backup behind it.

Can Claude close my books automatically?

No, and that is by design. /close-month reconciles and flags. It does not post the close on its own. You clear every flag and approve the result. It is a checklist helper, not an auto-close.

Should I run /month-heads-up before I close?

It is a good pairing. /month-heads-up looks forward at the next 30 days of cash and flags your tightest week. /close-month looks back and reconciles the month you just finished. Run the heads-up first, so you are not surprised by a cash crunch right after the books are done.

Can I run the close on a schedule?

Yes. Run /close-month once, type /schedule inside the task, and set it to repeat monthly. Scheduled tasks run with no device online. Anything that needs your sign-off waits for you the next time you open Claude.

Does it remember last month's fixes?

Only if you set it up to. A bare "work in a folder" session starts fresh each time. A Project carries files, instructions, and memory across sessions. Put a recurring client in a Project so your corrections stick.

How is this different from Growthy for many clients?

/close-month is a workflow tool: one folder, one session at a time. Growthy is built to run many client books at once. It connects to QuickBooks Online, uses pattern learning per client, and runs ten to thirty-plus client books with an audit trail. Both ask you to review before anything posts. Growthy just carries the volume Cowork was not built for.

Growthy is bookkeeping software, not a CPA firm. This content is educational, not professional advice.