Bank Statement MCP
Import a bank CSV export, categorise it, summarise per currency, find recurring charges and reconcile against your expense ledger. Everything stays on your machine.
Documentation
mcp-bank-statement
MCP server for reading bank statement CSV exports to read, categorise and reconcile transactions. Bank CSV exports categorised, summarised and reconciled with expenses.
Works with Claude Desktop, Claude Code, Cursor and any Model Context Protocol client. Runs on your own machine, or hosted with no install.
Product page: https://mcp.zovo.one/s/bank-statement — what it does, the tools it exposes, and a live token endpoint.
Install
Hosted, nothing to install. Get a token from https://mcp.zovo.one/mcp/connect (the connect page) or https://mcp.zovo.one/mcp/token (the same token as JSON); a free anonymous one is issued on the spot and a Pro key works the same way. Then point an MCP client at https://mcp.zovo.one/mcp/bank-statement over streamable-http and send the token as Authorization: Bearer <token>.
If your client cannot set headers, put the token in the path instead: https://mcp.zovo.one/mcp/bank-statement/t/<token>. Both forms work. The bare URL with no token answers 401 on tools/call, so the token is not optional.
Claude Desktop, one click. Download bank-statement.mcpb from the latest release and double-click it.
From source. The mirror is self-contained: every @theluckystrike/* dependency is vendored, so a fresh clone builds with no extra setup.
git clone https://github.com/theluckystrike/mcp-bank-statement.git
cd mcp-bank-statement
npm install && npm run build
Then point your client at the built entry point:
{
"mcpServers": {
"bank-statement": {
"command": "node",
"args": ["/absolute/path/to/mcp-bank-statement/dist/index.js"]
}
}
}
@theluckystrike/mcp-bank-statementis not published on npm yet, so annpx -y @theluckystrike/mcp-bank-statementcommand will fail. The three paths above are the working ones and each is exercised by CI.

Read-only mirror of mcp-servers/servers/bank-statement. See MIRROR.md.
In the official MCP Registry (io.github.theluckystrike/bank-statement-csv-categorize-reconcile-ledger).
Export the CSV from your bank, say "import this", and the month is readable. This MCP server turns a bank export into a local ledger: it finds the header row under whatever preamble the bank prints above it, works out which column is the date, which is the money and which way the money went, and reads amounts in the file's own locale, so 1 234,56 from mBank and 1,234.56 from a US bank both become the same number. A debit is stored negative and a credit positive, once, at import, so nothing downstream has to guess. Re-importing the same export adds nothing: every line carries a hash of its date, amount, currency and description, with an occurrence index, so two identical coffees on one day stay two transactions while a second import of the same file stays zero. Your rules categorise transactions as they arrive, summaries report money in, money out and the net per currency and never mix currencies, recurring_detect finds the subscriptions and what each costs per year, and reconcile_expenses matches bank debits against the receipts in mcp-expense-tracker so you can see which debits have no receipt and which receipts never reached the bank. Everything is a plain JSON file on your own machine; nothing is uploaded anywhere.
Turn a bank CSV into a categorised, reconciled month in chat, no accounting SaaS required.
Bank profiles: Revolut, Wise, mBank, PKO BP, ING, N26, and a generic reader that works from the headers alone.
File shapes it reads without being asked: UTF-8 and UTF-16 (either byte order, as Excel on Windows saves it),
CRLF, LF and CR-only line endings, quoted fields holding commas, quotes and newlines, 1.234,56 and 1,234.56,
accounting parentheses (12.50) and a trailing minus 12.50- (both negative), and a mixture of currencies inside
one file. A cell holding - or nothing is skipped and named in skipped_lines rather than read as zero. When the
date column is ambiguous (03/04/2026 with no value above 12 anywhere), the assumed order is stated in the
response instead of silently picking one.
60-second install
npm publish for @theluckystrike/mcp-bank-statement is pending. Until then, the .mcpb one-click bundle or a
clone+build is the working path, both are verified below.
One-click (.mcpb): download bank-statement.mcpb from the latest release and double-click it in Claude Desktop:
https://github.com/theluckystrike/mcp-servers/releases/latest
(claude_desktop_config.json):
{
"mcpServers": {
"bank-statement": {
"command": "npx",
"args": ["-y", "@theluckystrike/mcp-bank-statement"]
}
}
}
Claude Code:
claude mcp add bank-statement -- npx -y @theluckystrike/mcp-bank-statement
(.cursor/mcp.json):
{
"mcpServers": {
"bank-statement": {
"command": "npx",
"args": ["-y", "@theluckystrike/mcp-bank-statement"]
}
}
}
The npx form above starts working the moment the package is published. Until then, use the .mcpb bundle above, or
build from source with exactly these three commands:
git clone https://github.com/theluckystrike/mcp-servers.git && cd mcp-servers
npm install
npm run build -w packages/mcp-license -w servers/spreadsheet -w servers/bank-statement
Then point your client's command at node with one arg: the absolute path to servers/bank-statement/dist/index.js.
To run in Pro mode set MCP_LICENSE_KEY in the same config block, or call license_activate once with your key.
Tools
| Tool | What it does |
|---|---|
statement_import | Read a bank CSV into the ledger. Detects the header row under any preamble, the date, description, amount (or debit and credit) and currency columns, the date order and the number locale. Reports what it detected, what it stored, how many lines were duplicates of what is already there, and every line it skipped with the reason |
transactions_list | List transactions in a date range, filtered by account, category or uncategorized, with totals per currency |
transactions_search | Find transactions whose description, counterparty, category or account contains the query. Substring only, never a regex |
category_rules | Set or list the rules that categorise transactions. An empty match is refused, because it would stamp one category on the whole ledger. A plain match is a case-insensitive substring; regex: true compiles the pattern only if it cannot backtrack exponentially, and (a+)+ is refused and used as a substring. Setting rules re-applies them to stored transactions |
transaction_categorize | Set the category on transactions by id. Nothing changes unless every id exists |
statement_summary | Money in, money out and the net for a range, grouped by category, month, account or counterparty, always per currency |
reconcile_expenses | Match bank debits against the receipts in mcp-expense-tracker: same currency, same amount, within a date window. Reports matches, bank lines with no receipt, and receipts that never reached the bank. The expense ledger is only ever read |
recurring_detect | Find subscriptions and recurring charges: the same counterparty, an amount that barely moves, a steady cadence. Reports the cadence, the typical amount, the next expected date and the annualised cost. Two charges are one interval, so the yearly cost is withheld (cadence_confirmed: false) until a third charge confirms it |
statement_export | Write a range to csv or json and return the path. Written atomically, so a failure never leaves a half file. Replacing an existing file is allowed and reported (overwrote_existing_file) |
accounts_list | The imported accounts, with bank, currencies, transaction count, date range and last balance |
license_status | Show free or Pro mode |
license_activate | Activate a Pro key (verified offline) |
Resource bank://month returns this month's money in, money out and net, grouped by category, per currency.
Prompt monthly_review drives the whole month-end pass: totals, uncategorised lines, subscriptions, reconciliation.
Free vs Pro
| Free | Pro | |
|---|---|---|
| Accounts | 2 | unlimited |
| History read back | last 12 months | all of it |
| Category rules | 5 | unlimited |
| Import, list, search, summarise | yes | yes |
reconcile_expenses | yes, 31 days at a time | any date range |
recurring_detect | yes, last 3 months, 5 charges | full history, every charge |
statement_export | no | yes |
Import always stores every row, whatever the tier: silently dropping lines on the way in would make the ledger disagree with the bank. The free limit is on what is read back.
Get Pro: https://mcp.zovo.one/buy/bank-statement, $19 one-time for this server, $39 for every server, lifetime. Keys are verified offline; nothing is sent anywhere.
Pairs with
- mcp-expense-tracker, log the receipts and the mileage;
reconcile_expensesreads that ledger and tells you which bank debits have no receipt behind them. - mcp-spreadsheet, this server parses your bank CSV with the spreadsheet server's own RFC 4180
reader and locale number parser, so
1 234,56means the same thing in both. Use it to slice an exported range. - mcp-invoice, the income side: raise the invoice, then watch the payment land in
statement_summarygrouped by counterparty.
Privacy
Everything stays on your machine. Statements are parsed locally and stored in a plain JSON file under
${XDG_DATA_HOME:-~/.local/share}/mcp-servers/bank-statement/. The server makes no network calls at all, not for
licensing, not for parsing, not for anything. Deleting that directory resets it.
Built by theluckystrike. Support: support@zovo.one
Use these docs as an MCP server
Any MCP client (Claude, Cursor, Windsurf, VS Code) can read this repository's documentation directly via GitMCP — no install:
- Docs MCP URL: https://gitmcp.io/theluckystrike/mcp-bank-statement