holco-fec-controls
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., "@holco-fec-controlsanalyze this FEC text for balance and duplicate-line errors"
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.
HOLCO FEC Controls
Deterministic, explainable controls for synthetic French accounting export (FEC-like) data, exposed as a local MCP server for Claude and other compatible clients.
Deterministic when possible. AI when necessary. Human when accountable.
This repository is a clean-room public example. It contains no production code, customer data, credentials or internal HOLCO endpoints. It is not tax advice and does not certify legal or fiscal compliance.
Controls
required column presence;
parseable
YYYYMMDDdates;decimal debit and credit amounts;
debit/credit balance for every journal entry;
duplicate-line detection.
Every finding contains a stable code, severity, line or entry reference and a human-readable explanation. The AI client presents the result; it does not silently change the deterministic verdict.
Related MCP server: polyaccounts-demo
Control contract
The server separates planning from execution:
plan_fec_controldeclares controls, assumptions and decisions that require human input;analyze_fec_textruns every applicable deterministic control;the result binds evidence to the source with SHA-256 hashes and locators, without echoing accounting rows;
every planned control is reported as
PASS,REVIEW,FAILorNOT_RUN.
Business plausibility is deliberately NOT_RUN until an entity-approved policy
or benchmark is provided. A generic model opinion cannot turn a failed
accounting invariant into a pass. Consequently, a technically clean file has a
deterministic_outcome of PASS but remains globally REVIEW while the
declared business check is not run.
Run the tests
python -m pip install -e '.[mcp]'
python -m unittest discover -s tests -vConnect to Claude Desktop
After installation, add this local server to your Claude Desktop MCP configuration:
{
"mcpServers": {
"holco-fec-controls": {
"command": "holco-fec-mcp"
}
}
}Restart Claude Desktop, then ask it to call analyze_fec_text with FEC content.
The tool accepts text rather than a filesystem path, preventing unrestricted
file reads by the server. Keep real accounting exports within an environment
whose confidentiality rules you control.
MCP tools
list_fec_controls: returns the public control catalogue.plan_fec_control: previews scope, assumptions and human decisions.analyze_fec_text: analyses tab-separated content and returns structured results, evidence hashes, completeness and the stop reason.
See examples/synthetic_fec.txt for a fictitious
balanced entry.
Normative anchoring
The FEC (fichier des écritures comptables) is defined by article A.47 A-1 of
the French Livre des procédures fiscales and the arrêté of 29 July 2013: 18
normative fields in a fixed order, YYYYMMDD dates, and a normalised file
naming scheme. The DGFiP publishes an open-source structural checker, "Test
Compta Demat" (CeCILL licence, official DGFiP GitHub repository); article
1729 D of the CGI defines the sanction for a non-compliant file.
The controls in this repository target a subset of that referential on
synthetic data. Like the DGFiP checker itself, a green result is not an
attestation of conformity and does not bind anyone: it is a reproducible
technical statement about the supplied bytes. The full doctrine mapping of
the HOLCO control framework is maintained in the companion repository
holco-finance-controls
(DOCTRINE.md).
Licence
MIT. Copyright © 2026 HOLCO INVEST.
This server cannot be deployed
Maintenance
Related MCP Connectors
Your audit methodology inside Claude: findings, risks, controls and workpapers in your team format.
Read-only finance and operations controls for AI agents with evidence and safe next actions.
Connect Claude or Cursor to books, invoices, bills, payroll, and sealed closes.
Check AI work against requirements and return structured verdicts, findings, and repair steps.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables AI assistants to perform financial analysis, budget forecasting, compliance checks, expense categorization, and risk assessment, returning structured JSON with audit-ready governance receipts.567 npm1Business Source 1.1
- AlicenseNot gradedqualityAmaintenanceEnables users to query exact balances, expense changes, trial balances, and traceable ledger evidence from a synthetic company through read-only MCP tools and review prompts. It provides deterministic, integer-arithmetic accounting data for supervised evaluation without requiring an account or network connection.MIT
- AlicenseCqualityAmaintenanceEnables auditing of extracted balance sheets, income statements, and cash flow reports by reconstructing complex document tables and verifying mathematical consistency and formula balancing. It also performs cross-reference resolution and sensitive PII redaction, returning structured matrices and validation telemetry over MCP to clients like Claude Desktop, Cursor, and Windsurf.47MIT
- AlicenseAqualityCmaintenanceEnables a language model to run four deterministic data checks it cannot argue with — column-contract validation with an explicit quarantine list, arithmetic reconciliation of opening/movements/closing, distinct entity counting against row counts, and duplicate detection returned for human review — each answering with structured findings and row/column locators. Failures come back as results with reasons, so the model reports exactly which rows and columns are broken rather than asserting the data is clean.4MIT