mbank-parser-mcp
This MCP server lets you locally parse and analyze mBank CSV operation exports through three AI-callable tools, with no network access.
read_statement_header – preview client name, date range, accounts, inflow/outflow totals, and operation count without loading operation rows.
list_operations – parse and filter operations by date range, category, account, description substring, amount range, and limit results.
summarize_operations – aggregate operations by category, account, or month, returning counts, inflow, outflow, net per bucket, and grand totals.
Handles Polish-locale number formats (space thousands, comma decimals,
PLNsuffix) and returns signed float amounts.Works entirely offline on local files; only dependency is the MCP SDK.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mbank-parser-mcp Show me all card payments over 500 PLN in Q2."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mbank-parser-mcp
Local MCP server that parses mBank CSV operation exports — the exact file you can download from the mBank web app under "Historia" → "Eksportuj do CSV". Runs entirely offline. Nothing leaves your machine.
Part of the honest-mcp family of small, auditable, local-first MCP servers.
Why
Anyone using mBank in Poland — private, JDG, sp. z o.o. — does the same monthly ritual:
Log into mBank, download the CSV.
Open in Excel, squint at Polish-locale amounts (space-thousands, comma-decimals,
PLNsuffix).Filter by date, sum by category for VAT / income tax reconciliation.
This server automates step 2 and 3 through your AI. You keep your bank data on disk; the AI just calls a tool that reads and filters it locally.
Zero network calls. The requirements.txt is one line — the MCP SDK. No dependency footprint to audit.
Related MCP server: monter-mcp
Features
Three tools:
read_statement_header— quick preview of a file: client name, date range, listed accounts, inflow/outflow totals, operation count. No operation rows loaded.list_operations— parse + filter operations. Combine any of:date_from,date_to,category,account,contains(description substring),min_amount,max_amount,limit.summarize_operations— aggregate bycategory,account, ormonth. Returns count, inflow, outflow, net per bucket plus grand totals.
Amounts come back as signed floats (negative = expense), currency separate. Polish-locale numbers (-3 283,33 PLN) and non-breaking-space thousand separators are handled.
Supported files
✅ CSV export from mBank web banking ("Historia operacji" → "Eksportuj do CSV")
❌ PDF statements — mBank's layout varies too much per statement type for reliable parsing. Skipped intentionally.
Requirements
Python 3.10+
Setup
git clone https://github.com/bartosz-kuc/honest-mbank-parser-mcp.git
cd honest-mbank-parser-mcp
python3 -m venv venv
./venv/bin/pip install -r requirements.txtOn Windows use python instead of python3, and venv\Scripts\pip and venv\Scripts\python instead of venv/bin/pip and venv/bin/python (here and in the configs below).
Register with Claude Code:
claude mcp add mbank /absolute/path/to/venv/bin/python /absolute/path/to/server.pyClaude Desktop claude_desktop_config.json:
{
"mcpServers": {
"mbank": {
"command": "/absolute/path/to/venv/bin/python",
"args": ["/absolute/path/to/server.py"]
}
}
}Example usage
"How much did I spend on 'Ubezpieczenia' in Q2?"
summarize_operations(file_path=".../lista_operacji.csv", group_by="category", date_from="2026-04-01", date_to="2026-06-30") → totals per category, including "Ubezpieczenia".
"List all incoming transfers over 10 000 PLN this year."
list_operations(file_path=".../lista_operacji.csv", min_amount=10000, date_from="2026-01-01")
"Show me every card purchase mentioning Anthropic."
list_operations(file_path=".../lista_operacji.csv", contains="ANTHROPIC")
Data flow
Your AI client
↕ MCP stdio
This server (Python, on your machine)
↕ local filesystem
Your mBank CSVNo HTTP, no external service — the server has no requests or httpx dependency at all.
Author
Bartosz Kuć — Warsaw-based developer, JDG owner running skanfirmy.pl.
GitHub: https://github.com/bartosz-kuc
Email: firma@bartosza.pl
Consulting
Available for consulting on Polish tax and business integrations (KSeF, GUS/NFZ/GIOŚ APIs, mBank data), MCP server design, and AI-assisted tooling for JDGs and small teams. See skanfirmy.pl/uslugi for productized packages (audit 3k PLN, setup 8-15k PLN, retainer 2-4k PLN/mo), or reach out via email.
License
MIT — see LICENSE.
Related
Part of the honest-mcp family — see the family index.
Available Tools
3 toolslist_operationsA
Parse an mBank CSV operations export and return the operation rows (date, description, account, category, signed amount, currency). All filters are optional; combine them freely. Amounts are already parsed to floats (negative = expense). For big statements consider a limit to keep response size in check.
Inputs: file_path (required), limit, account, date_to, +6 more.
read_statement_headerA
Read only the header block of an mBank CSV export — client name, date range, listed accounts, inflow/outflow totals — without loading the operation rows. Fast preview to confirm you have the right file.
Inputs: file_path (required).
summarize_operationsA
Aggregate operations by category, by account, or by month. Returns count, total inflow, total outflow, net per bucket, plus the grand totals over the filtered set. Ideal for building a monthly expense report or checking VAT-relevant categories.
Inputs: file_path (required), account, date_to, category, +5 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 3 tools
Each tool targets a distinct stage of working with an mBank CSV: header preview, operation rows, and aggregations. There is no meaningful overlap between reading metadata, listing transactions, and summarizing them.
All tool names follow a consistent verb_noun pattern: read_statement_header, list_operations, summarize_operations. The verbs clearly indicate the action and the nouns clearly indicate the target resource.
Three tools is well-scoped for a focused CSV parser server. Each tool covers a necessary capability without redundancy or feature bloat.
For the stated purpose of parsing mBank CSV exports, the surface is complete: preview the header, retrieve operations with filtering, and summarize them. No obvious missing operation is implied by the domain.
Maintenance
Related MCP Connectors
Track expenses from your AI chat: log, categorize, monthly CSV reports. All data is local.
Chat with your bank data: balances, transactions, budgets, bills. Reads only, never moves money.
Import bank CSV exports, categorise transactions, total them by category or month, match expenses.
Ask Claude about your income and spending, and import bank statements. No bank login.
Related MCP Servers
- AlicenseBqualityCmaintenanceEnables local-first personal finance management through deterministic tools for importing, categorizing, and analyzing bank transactions.36MIT
- AlicenseNot gradedqualityDmaintenanceEnables local analysis of personal finance data from a Monter JSON export, answering questions about spending, subscriptions, debts, and more via MCP-capable assistants without any cloud servers.MIT
- AlicenseNot gradedqualityBmaintenanceEnables querying your own local bank and credit card statement files (CSV or PDF) in plain language, so you can discover forgotten recurring subscriptions, compare spending across months and quarters, and look up transactions without any bank logins or network calls.MIT
- AlicenseBqualityCmaintenanceEnables AI assistants to analyze a local bank statement CSV through seven tools covering spending by category, top merchants, recurring subscriptions, unusual or duplicate charges, end-of-month forecasts, and transaction search, plus a monthly review prompt. CSV files stay on the user's machine, with only aggregated results and matched transactions exposed to the assistant.7MIT