Yes, Claude can reconcile your bank account. Connected to LedgerMCP, it compares your recorded transactions to the bank statement, ties the two together to the cent, and saves the result as proof the period settled. This is not Claude reading a PDF and guessing. It is Claude calling a real reconciliation tool that either confirms your books match the bank or hands back a short list of what to fix. Here is how it works.
How does Claude reconcile an account?
Claude reconciles by calling one tool, reconcile_account, with the account and the statement it is checking against. The tool ties your book to the statement two ways: by the ending balance printed on the statement, or by a full line-by-line worksheet when you have every cleared item. Claude picks whichever matches the source you handed it. If the books agree with the bank, the tool returns a difference of zero and records the tie-out. If they disagree, it returns the exact gap and points at the rows most likely to explain it.
The important part is that Claude is not the one deciding whether the numbers balance. The ledger computes that. Claude drives the tool, reads the result, and does the follow-up work. You can read the full manual method in our guide to reconciling a bank account step by step, then see how the same steps collapse into one request here.
Does Claude save the reconciliation as proof?
Yes, and this is what separates a real reconciliation from a chat about your finances. Every time Claude ties a period out, LedgerMCP persists it as a stored reconciliation object: which account, which period, what balance, and when it settled. You can ask Claude to list past reconciliations and see exactly which months are proven.
A settled period stays settled. If someone later posts a back-dated entry that disturbs a month you already closed, a full books check re-derives every reconciliation and names the ones that no longer tie. So the proof is not a screenshot that goes stale. It is a live check the ledger keeps honest for you, which is why closing and locking each month, covered in a month-end close with Claude, actually holds.
What happens when it does not tie? An example
Say your books show a bank balance of $12,384.10 and the statement ends at $12,341.60. Claude runs reconcile_account and the tool reports the account is off by $42.50. Instead of asking you to hunt through the whole month, the tool returns a ranked list of suspects. Claude works it in order:
- A missing transaction equal to the gap. The most common cause. A $42.50 charge cleared the bank but was never recorded in the books. Claude adds it, categorizes it, and the difference goes to zero.
- A duplicate. The same $42.50 payment got recorded twice. Claude reverses the extra one.
- A transposed amount. A charge entered as $42.50 that cleared at $85.00, or a $24.50 typed as $42.50. Claude corrects the figure to match the bank.
Because postings are immutable, Claude never quietly overwrites your data. A fix is a linked reversal plus a clean re-post, so the trail shows what changed and why. Once the offending row is handled, Claude re-runs the reconciliation and confirms the difference is now zero.
Why is Claude good at chasing the difference?
Reconciling by hand is tedious for one reason: when you are off by $42.50, you have to scan dozens of rows looking for the culprit. That scan is exactly the kind of pattern work a model is fast at, and the tool makes it faster still by returning the suspects rather than the whole statement. Claude checks whether any single transaction equals the gap, whether two amounts sum to it (a transposition), and whether a merchant appears twice on the same day. It reasons over the candidates the ledger surfaces instead of you eyeballing a page.
This is drafting-and-approving, not blind automation. Claude proposes the fix, and you can preview the exact entry before it posts. Nothing hits your books without being reviewable, and anything posted has a one-click reverse.
What do you need to give Claude to reconcile?
One outside source of truth. That can be the ending balance from a PDF statement, a full line-by-line statement, or a connected bank feed. Feeds start at $9/mo pooled across your books, but you do not need one: the ending balance off a paper statement is enough for Claude to tie the period out. You bring the statement; Claude does the reconciliation. Everything else, the accounts, the balanced entries, and the stored proof, lives in LedgerMCP, the double-entry books that Claude keeps for you through the reconciliation tool and the rest of the ledger contract.
For the mechanics of the tie-out itself, the worksheet, the ending-balance method, and what usually breaks a reconciliation, see LedgerMCP's bank reconciliation feature. It is the same engine Claude drives when you ask it to reconcile.
Quick answers
Can Claude actually reconcile a bank account?
Yes. Connected to LedgerMCP, Claude calls the reconcile_account tool, compares your recorded transactions to the statement, and saves the tie-out as a stored object. It is not summarizing a PDF; it is running the same reconciliation an accountant would, then persisting the proof that the period tied.
What happens when the reconciliation does not tie?
The tool returns the difference and a ranked list of likely culprits: a missing transaction equal to the gap, a duplicate, or a transposed amount. Claude investigates the suspects, fixes or adds the offending row, and re-runs until the difference is zero.
Does Claude need my bank statement?
Yes, it needs an outside source of truth. You can reconcile to the ending balance printed on a PDF statement, to a full line-by-line statement, or against a live bank feed. Claude works from whichever one you give it.
Is the reconciliation saved as proof?
Yes. Every tie-out persists as a reconciliation object you can list later, so you know which periods settled and when. If a back-dated entry later disturbs a closed period, a books check flags that the reconciliation no longer ties.
