mt940 parses MT940 bank statement files into smart, fully typed Python collections you can iterate, aggregate and serialize. It has no runtime dependencies, ships type information (py.typed), and copes with the quirks of many real-world banks.

import mt940

transactions = mt940.parse('statement.sta')
for transaction in transactions:
    print(transaction.data['date'], transaction.data['amount'])

• Zero runtime dependencies — pure standard library.
• Fully typed — ships py.typed; checked under pyright, mypy and pyrefly.
• Battle-tested — 100% test coverage against fixtures from many banks.
• Smart models — amounts, balances and dates come back as rich Python objects, not raw strings.
• JSON-ready — a single encoder serializes a whole statement.
• Extensible — opt-in tags and pre/post processors for bank-specific formats.
• Modern Python — supports 3.10 through 3.13.

pip install mt-940

Using uv:

uv add mt-940

Requires Python 3.10 or newer.

parse() accepts a filename, an open file handle, or the raw str/bytes contents, and returns a Transactions collection:

import mt940
import pprint

transactions = mt940.parse('mt940_tests/jejik/abnamro.sta')

# Statement-level data (balances, account, ...) lives on the collection:
pprint.pprint(transactions.data)

# Iterate the individual transactions:
for transaction in transactions:
    print(transaction.data['date'], transaction.data['amount'])
    pprint.pprint(transaction.data)

Each Transaction exposes a data dictionary with the parsed fields. Which fields are present depends on the source bank and the tags in the file.

Statement-level balances live on the Transactions object's data, not on the individual transactions — this works even for files with no transactions at all:

import mt940

transactions = mt940.parse('statement.sta')
print(transactions.data['final_opening_balance'])
print(transactions.data['final_closing_balance'])
print(transactions.data['available_balance'])

import mt940
import pprint

mt940.tags.BalanceBase.scope = mt940.models.Transaction

# The currency has to be set manually when moving the BalanceBase scope to
# Transaction.
transactions = mt940.models.Transactions(processors=dict(
    pre_statement=[mt940.processors.add_currency_pre_processor('EUR')],
))

with open('mt940_tests/jejik/abnamro.sta') as fh:
    transactions.parse(fh.read())

for transaction in transactions:
    pprint.pprint(transaction.data)

A single parse() merges everything into one Transactions and keeps only the last block's statement-level data (e.g. balances). For files that concatenate several statements (including balance-only blocks), use parse_statements(), which splits on :20: boundaries and returns one Transactions per statement, each with its own balances:

import mt940

# src may be a filename, a file handle or the raw data, just like parse()
for statement in mt940.parse_statements('statements.sta'):
    print(statement.data['final_opening_balance'])
    print(statement.data['final_closing_balance'])

import json
import mt940

transactions = mt940.parse('statement.sta')
print(json.dumps(transactions, indent=4, cls=mt940.JSONEncoder))

By default a new transaction is started only on the :61: statement tag. Some banks delimit transactions differently — for example by repeating the :20: transaction reference per block. Because changing the default grouping would break existing users, this behaviour is opt-in: pass transaction_boundary (an iterable of tag slugs) to start a new transaction on those tags too.

import mt940

# Each `:20:` (transaction_reference_number) starts its own transaction:
transactions = mt940.parse(
    'statement.sta', transaction_boundary={'transaction_reference_number'}
)

The same option is accepted by mt940.models.Transactions(transaction_boundary=...).

Some banks (e.g. GLS / Atruvia) put a customer reference longer than the SWIFT 16-character cap on the :61: line, followed by the // bank reference. Relaxing the default would change how other banks (e.g. Rabobank) split same-line data, so this is handled by an opt-in StatementGLS tag:

import mt940

gls = mt940.tags.StatementGLS()
transactions = mt940.parse('statement.sta', tags={gls.id: gls})

(Longer supplementary details — issue #117, e.g. Wise — are handled by the default parser and need no opt-in.)

Tag 61 in ASN statements does not follow the SWIFT specification, so the opt-in StatementASNB tag is used:

import mt940
import pprint

tag = mt940.tags.StatementASNB()
transactions = mt940.models.Transactions(tags={tag.id: tag})

with open('mt940_tests/ASNB/mt940.txt') as fh:
    transactions.parse(fh.read())

pprint.pprint(transactions.data, sort_dicts=False)

Help is greatly appreciated. Please clone the develop branch and run tox before opening a pull request; CI checks linting (ruff), type-checking (pyright, mypy, pyrefly), the test suite (100% coverage required), and the documentation build.

git clone --branch develop https://github.com/WoLpH/mt940.git
cd mt940
uv sync
uv run tox          # run the full matrix
uv run tox -e py312 # or a single environment

• Documentation: https://mt940.readthedocs.io/
• Source: https://github.com/WoLpH/mt940
• Bug reports: https://github.com/WoLpH/mt940/issues
• Package: https://pypi.org/project/mt-940/
• License: BSD-3-Clause