We didn’t win, but we proved what we set out to prove
The Qonto x Anthropic MCP Hackathon wrapped up on July 13th: 54 projects submitted in just a few days, all built on the same ingredient — Qonto’s MCP connector, which lets Claude read and act on real banking data in real time.
We took part with Nexyzen Clearing Scout, a Claude Skill that finds working capital frozen in unpaid invoices and shows how much of it can be released without a single bank transfer, through debt clearing.
Ask one plain-language question about your cash position.
The skill:
- reads client and supplier invoices straight from Qonto;
- spots counterparties who are both a creditor and a debtor — an instant bilateral offset that needs no network at all;
- maps the wider network of debts across counterparties to find multilateral clearing cycles no single company could ever see from its own books alone;
- explains it all in plain language, and never sends anything anywhere without explicit user confirmation.
No number is guessed by the model: every figure comes from deterministic, auditable scripts — same input, same output, every time.
The 3-minute demo video and the pull request are public: demo video · PR on github.com/qonto/skills
The winning project, Argentier by Kevin Mameri, genuinely deserved it: it cross-references a company’s real spending against market rates and hands you a ready-to-send negotiation email. Well earned.
What we’re taking away is different, but just as meaningful: the hackathon format rewards skills you run on demand, inside a single Claude conversation. Our use case — multilateral clearing — by its very nature needs network-wide data that no single company, and no single session, can see on its own.
That’s not a shortcoming of the project: it’s proof that Nexyzen’s clearing engine can plug directly, via API, into existing banking infrastructure like Qonto’s — not just as an occasional skill, but as an always-on capability.
And that’s exactly what we set out to show: not just that Nexyzen works, but that it’s ready to be integrated.
Sincere thanks to the Qonto and Anthropic teams for putting this hackathon together, and to every other participant for the energy they brought to it.