MCP 服务器

bankstatementparser-mcp

Model Context Protocol server exposing the bankstatementparser library as first-class agent tools for reading bank statements.

README

<h1 align="center">bankstatementparser-mcp</h1>

Model Context Protocol server exposing the bankstatementparser library as first-class agent tools for reading bank statements.

Getting started

What is bankstatementparser-mcp? — the problem it solves
Install — PyPI, virtualenv, Docker
Quick start — register with Claude Desktop in 30 seconds

Library reference

Tools — the five tools, one resource, one prompt
Using the tools — call them in-process from Python

Operational

When not to use bankstatementparser-mcp — honest boundaries
Development — gates, make targets
Security — sandboxing posture
Documentation — examples, guides
Contributing — how to get changes in
License — Apache-2.0

What is bankstatementparser-mcp?

The Model Context Protocol (MCP) is an open standard that lets AI agents discover and call external tools in a uniform way. bankstatementparser-mcp is the MCP server that turns the bankstatementparser library into first-class agent tools — so an assistant can read, validate, and summarise bank statements in formats such as ISO 20022 CAMT.053, SWIFT MT940, OFX/QFX, and CSV directly from a conversation.

Every tool is a thin wrapper over the bankstatementparser parser core (create_parser, detect_statement_format), so the results behave identically to the CLI. Because an MCP client does not share the server's filesystem, the tools take inline statement content (plus a filename hint) and materialise it in a private temporary file for the duration of a single call. Tools return JSON-serialisable data.

Concern	How bankstatementparser-mcp handles it
Transport	stdio (FastMCP default); zero config beyond the client manifest
Input model	Inline content + filename hint; no shared filesystem required
Format fidelity	Tools delegate to `bankstatementparser`'s `create_parser` pipeline
Format detection	`detect_format` mirrors the library's `detect_statement_format`
Validation	`validate_statement` is a dry run that returns structured results
Isolation	Each call writes to a private temp file that is deleted on exit

Install

Channel	Command	Notes
PyPI	`pip install bankstatementparser-mcp`	Pulls in `bankstatementparser >= 0.0.9` + MCP SDK
Source	`git clone https://github.com/sebastienrousseau/bankstatementparser-mcp && cd bankstatementparser-mcp && poetry install`	For development
Docker (GHCR)	`docker pull ghcr.io/sebastienrousseau/bankstatementparser-mcp:latest`	Multi-arch (linux/amd64, linux/arm64); runs `bankstatementparser-mcp` over stdio

Requires Python 3.10 or later. Works on macOS, Linux, and Windows.

<details> <summary>Using an isolated virtual environment (recommended)</summary>

python -m venv venv
source venv/bin/activate        # macOS/Linux
venv\Scripts\activate           # Windows
python -m pip install -U bankstatementparser-mcp

</details>

Quick start

{
  "mcpServers": {
    "bankstatementparser": { "command": "bankstatementparser-mcp" }
  }
}

That's it. Restart the client and the tools are available to the agent.

The server speaks JSON-RPC over stdin/stdout — it is meant to be launched by an MCP client, not used interactively.

Tools

All tools delegate to the bankstatementparser parser core, so they behave identically to the library.

Tool	Purpose
`list_supported_formats`	List every bank statement format the parser can read
`detect_format`	Detect which statement format an inline payload is
`parse_statement`	Parse a statement into structured transactions plus a summary
`validate_statement`	Dry-run check whether a statement parses cleanly
`summarize_statement`	Return only the statement summary (no per-transaction rows)

Plus one resource and one prompt:

Kind	Name	Purpose
Resource	`bankstatementparser://formats`	Read-only catalogue of supported formats and their file extensions
Prompt	`analyze_statement`	Guided multi-step prompt that walks an agent through reading and reconciling a statement

Supported formats: camt (ISO 20022 CAMT.053, .xml), pain001 (ISO 20022 pain.001, .xml), csv (.csv), ofx (.ofx), qfx (.qfx), and mt940 (SWIFT MT940, .mt940 / .sta).

Using the tools

The tools are plain functions on the bankstatementparser_mcp.server module, so you can call them in-process:

from bankstatementparser_mcp.server import (
    detect_format,
    parse_statement,
    summarize_statement,
)

csv = (
    "date,description,amount,currency,balance\n"
    "2023-01-02,Salary,500.00,EUR,1500.00\n"
    "2023-01-03,Groceries,-40.50,EUR,1459.50\n"
)

# 1. Detect the format from the filename hint + content.
print(detect_format(csv, "statement.csv"))
# -> csv

# 2. Parse the statement into structured rows + a summary.
parsed = parse_statement(csv, "statement.csv")
print(parsed["transaction_count"], parsed["columns"])

# 3. Read just the opening/closing balances.
print(summarize_statement(csv, "statement.csv"))

The resource and prompt are plain functions too: formats_resource backs bankstatementparser://formats, and analyze_statement returns the guided multi-step prompt.

from bankstatementparser_mcp.server import (
    analyze_statement,
    formats_resource,
)

print(formats_resource())          # the supported-formats catalogue
print(analyze_statement("statement.csv"))  # the guided analysis prompt

See the examples/ folder for runnable walkthroughs, including 04_resource_and_prompt.py.

When not to use bankstatementparser-mcp

You're not driving an MCP-aware agent. Use the bankstatementparser CLI or library directly — it exposes the same surface with less indirection.
You need to parse files already on disk in bulk. The library's CLI reads paths directly and avoids the inline-content round-trip the MCP tools use.

Development

bankstatementparser-mcp uses Poetry and mise.

git clone https://github.com/sebastienrousseau/bankstatementparser-mcp.git
cd bankstatementparser-mcp
mise install
poetry install

A Makefile orchestrates the quality gates (kept in lockstep with CI):

Target	What it runs
`make check`	All gates (REQUIRED before commit)
`make test`	`pytest --cov=bankstatementparser_mcp --cov-branch --cov-fail-under=100`
`make lint`	`ruff check` + `black --check`
`make type-check`	`mypy --strict`
`make docs`	`interrogate --fail-under=100` (docstring coverage)

Current state (v0.0.11): 100% line + branch coverage against a 100% enforced floor, mypy --strict clean, interrogate 100%.

Security

No persistent filesystem writes from tools. Each call writes the inline content to a private temporary file that is deleted as soon as the call returns.
Validation failures from validate_statement are returned as structured {"is_valid": false, "error": ...} payloads — never as stack traces.
Dependencies are pinned via poetry.lock and audited by pip-audit and Bandit in CI.

To report a vulnerability, please use GitHub private vulnerability reporting rather than a public issue.

Documentation

Runnable examples: examples/
Release history: CHANGELOG.md
MCP specification: modelcontextprotocol.io

Contributing

Contributions are welcome — see the contributing instructions. Thanks to all the contributors who have helped build bankstatementparser-mcp.

License

Licensed under the Apache License, Version 2.0. Any contribution submitted for inclusion shall be licensed as above, without additional terms.

<a href="https://bankstatementparser.com">bankstatementparser.com</a> · <a href="https://pypi.org/project/bankstatementparser-mcp/">PyPI</a> · <a href="https://github.com/sebastienrousseau/bankstatementparser-mcp">GitHub</a>