Qonto Grant Scout

Your bank account knows what you invest in; public databases know who gets helped for it. This skill introduces them. Read-only, zero writes: it profiles, searches, and shortlists — it never claims eligibility and never fills an application.

Prerequisites

1. get_organization FIRST → legal form, sector (NAF code when present), address/region, creation date, accounts (list_transactions requires bank_account_id/iban). If the sector field is absent, infer it from counterparties and flow patterns — and say so, with lower confidence.
2. Country-aware: aid schemes are national/regional → cross-referencing is France only. For other Qonto countries (DE, ES, IT, AT, NL, BE, PT…), say so plainly and degrade to what stays valid everywhere: the spending profile and the generic aid families that exist across the EU — never cite a specific foreign scheme.
3. Datagouv MCP: optional, detected dynamically. Present → dataset search (search_datasets, get_dataset_info, query_resource_data). Absent → announce it and still deliver the spending profile + generic aid families with the official portals to explore (aides-entreprises.fr, bpifrance.fr, the region's economic-development site, France Num).

Workflow

1. Profile the company

From get_organization: legal form, sector/NAF, region (from address), company age, size proxy (incoming volume over the last 12 months). Build a one-paragraph profile the user can correct before anything else happens. Every later search inherits this profile.

2. Read the flows (12–24 months)

list_transactions per account, paginate per_page: "50". Classify debits into aid-relevant spending categories: training (training bodies, e-learning, certifications) · equipment/CAPEX · digital (software, SaaS, web) · R&D-like (prototyping, technical subcontracting, specialized tooling) · hiring/payroll trend · international (foreign counterparties, FX) · energy/vehicles/works. Use list_labels for the user's own categorization when available. Match card spending by emitted_at, not settled_at. Annualize each category and keep the evidence (n transactions, top counterparties, trend).

3. Extract "aidable" signals

A signal = category + annual amount + trend + evidence. Keep the significant ones: recurring training spend, an equipment purchase or a visible CAPEX ramp, payroll growth, export activity, energy/efficiency spending, sustained digital investment. Show the signals table before searching — this is the "your account says this about you" moment.