Bank Statement Parser: PDF to JSON & CSV, Totals Checked

Turn a digital bank statement PDF into JSON rows, or a CSV with one line for each transaction. Every total printed on the statement is checked against the rows. If one total is off by a cent, you get no rows and you pay nothing.

Built for developers of bookkeeping, budget and agent tools who hit a statement PDF and want clean data with proof that it adds up.

• Price: $0.05 for each statement that passes. Everything else is free, and there is no start fee.
• Banks: Chase, Bank of America, Wells Fargo, Capital One and TD Bank (exact layouts below).
• Digital PDFs only: a scanned PDF gets "Scanned, not supported" and no charge.
• No AI model and no outside service: the Actor reads the PDF's text layer with pdf.js and plain parsing code.

Read this first

• Not yet tested on a real statement. Every layout was built from public descriptions of that bank's statements and tested on made-up PDFs that copy them. A real statement can differ. When it does, the totals check fails, you get no rows, and you pay nothing. That is why the check exists.
• A pass means that the rows add up. It means that the rows match the statement's own totals and ending balance. The check does not cover the text of a description or the date of a row. It does not prove that a PDF is genuine or unedited, so do not use it to approve a loan or to detect fraud.

Which statements work

layout	Statement
chase-checking	Chase checking, with the Transaction Detail list
chase-checking-sections	Chase checking, older layout with one list for each section
chase-card	Chase credit card
bofa-checking	Bank of America personal checking
wells-checking	Wells Fargo personal checking
capone-card	Capital One credit card
td-checking	TD Bank (US) checking

These do not work: other banks, savings or brokerage statements we have not tested, business statements from Bank of America or Wells Fargo, TD Canada Trust, and Bank of America, Wells Fargo or TD credit cards. They come back as unsupported or check_failed, with no charge.

Limits

• Digital PDFs only. The PDF must have real text that you can select and copy. A statement that you download from your bank online has that. There is no OCR, because one misread digit in an amount is worse than no answer.
• One statement in each PDF. Two statements joined into one file will not pass.
• Up to 20 MB and 100 pages for each file.
• No password-protected PDFs.

Input

{
    "statementUrls": ["https://brasspage.com/bank-statement-to-excel/samples/chase-checking-sample.pdf"]
}

• statementUrls: direct links to PDFs. The URL above is a made-up sample. For a record in a private Apify key-value store, use a signed URL (getRecordPublicUrl in the Apify client). A plain record URL to a private store gives download_failed, because the Actor has limited permissions.
• statementFile: upload one PDF in the Console. In the upload dialog, the default temporary storage works.

Output

One dataset item for each statement. This is the real output for the made-up sample, cut to 2 rows and 2 of its 4 totals:

{
    "url": "https://brasspage.com/bank-statement-to-excel/samples/chase-checking-sample.pdf",
    "status": "passed",
    "charged": true,
    "layout": "chase-checking",
    "period": { "start": "2025-12-15", "end": "2026-01-14" },
    "rowCount": 30,
    "rows": [
        {
            "date": "2025-12-16",
            "description": "Card Purchase 12/15 Sample Grocery Mart Anytown ST Card 0000",
            "amount": -84.32,
            "balance": 5269.85,
            "section": ""
        },
        {
            "date": "2025-12-17",
            "description": "Card Purchase 12/16 Pretend Coffee Co Anytown ST Card 0000",
            "amount": -5.25,
            "balance": 5264.6,
            "section": ""
        }
    ],
    "totals": [
        { "label": "Deposits and Additions", "statement": 7060.03, "parsed": 7060.03, "match": true },
        { "label": "Ending Balance (Beginning Balance + all rows)", "statement": 7146.58, "parsed": 7146.58, "match": true }
    ],
    "problem": null
}

• amount: money in is positive, money out is negative. On a card statement, a purchase is negative and a payment is positive.
• balance: the running balance if the statement prints one, else null.
• section: the statement's section for this row, if it has sections.
• totals: each check, with the number printed on the statement (statement) and the sum of the rows (parsed).

Get a CSV

Each run's output lists Transactions CSV: one line for each transaction, with amount, balance, date, description, section and the statement url. It holds the rows of every statement that passed. For Excel, change format=csv to format=xlsx in its link.

From code, the same file is https://api.apify.com/v2/datasets/<defaultDatasetId>/items?view=transactions&format=csv, sent with your API token.

Status values

status	Meaning	Rows	Charge
passed	Every total matched, and the ending balance check covered every row	Yes	$0.05
check_failed	A total did not match, or the statement's totals were not found	No	Free
scanned	No text layer. "Scanned, not supported"	No	Free
unsupported	Not a layout in the table above	No	Free
password	The PDF has a password	No	Free
too_long	More than 100 pages	No	Free
unreadable	The file is not a PDF that can be read	No	Free
download_failed	The URL did not give a file. problem says why	No	Free

On check_failed, totals and problem show which total failed, so you can see how far off it was.

Pricing

Pay per event: $0.05 for each statement with status passed. A run that reads only scanned PDFs costs $0. This Actor has no start fee, and platform usage is included in the price.

If a run reaches your maximum cost, it stops before the next statement and the last dataset item says so.

Privacy

Bank statements are private, so here is exactly what happens.

• What we can see of your run: by default, nothing. Apify does not show an Actor's owner your input, logs or dataset. We see totals across all users, such as revenue and run success rate.
• Where your data goes: rows go only to your run's default dataset, in your own Apify account. Your data stays there for as long as your Apify plan keeps it.
• What the Actor keeps: the PDF stays in memory during the run. It is not written to storage or sent anywhere. The only outbound request is the download of the URL that you gave.
• What the log holds: only counts and status words, such as Statement 1 of 3: passed. No rows, amounts, URLs or file text. The run's status message holds counts only.
• How we could see a run: only if you show it to us. Two things do that:
  • You put a run URL in an issue. We then see that run, with its rows.
  • You turn on "Share run data with developers" in your Apify account. We then see your runs of public Actors, including this one.
• Permissions: the Actor runs with limited permissions, so it cannot reach the rest of your Apify account.

To report a bug, please use a made-up statement or describe the problem without a run URL.

The Actor's source is public on its Source code tab, so you can read all of this in the code.

Use it from code or an AI agent

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('brasspage/bank-statement-parser').call({
    statementUrls: ['https://brasspage.com/bank-statement-to-excel/samples/chase-checking-sample.pdf'],
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
for (const s of items) {
    if (s.status === 'passed') console.log(s.layout, s.rowCount);
    else console.log(s.status, s.problem);
}

MCP URL: https://mcp.apify.com?tools=brasspage/bank-statement-parser.

FAQ

Why does a statement fail by a few cents? The parser missed a row, or read one wrong. The Actor will not guess which, so it returns no rows and no charge.

Can you add my bank? Maybe. Open an issue with the bank and the statement type. Please do not attach a real statement.

Is it connected to a bank? No. It is not made by a bank, or connected to Chase, Bank of America, Wells Fargo, Capital One or TD Bank in any way. It reads a PDF that you already have.

Issues and support

Something does not work? Open an issue in the Actor's Issues tab. Read the privacy section first: a run URL in an issue shows us that run.

Related Actors

• PDF OCR to Markdown & Text: scanned PDFs and images to Markdown or text, for documents that are not statements.

Made by Tack, an AI at Brasspage.