Forecast my cash for the next 13 weeks

Using the Qonto MCP, forecast my cash week by week for the next 13 weeks, based on how my clients actually pay and on what my account really spends, show me my lowest week and tell me what to do now to avoid a squeeze.

Steps

1. Starting point. get_organization for country and legal form, list_bank_accounts for the balance of each operating account. Savings or reserve accounts are listed apart and excluded unless the user includes them. Week 1 starts on the Monday of the current week.

2. Learn the history (6 to 12 months). list_transactions per account, per_page: "25", paginated: all credits, and debits filtered on operation_type: ["direct_debit", "transfer"]; list_cash_flow_categories for category names. Do not page every card payment: card and fee outflows over the last 12 weeks = credits - change in balance - direct debits - transfers out, with the balances read from settled_balance (one call, per_page: "1", sort_by: "settled_at:desc", settled_at_to = the start date). Credits from accounts in the founder's own name are owner contributions, not revenue, even when categorised as sales: show them apart and keep them out of the inflow series. Normalise counterparties (uppercase, strip SEPA and card prefixes, references and legal suffixes). A series is recurring when it has at least 3 monthly or 2 quarterly occurrences, regular intervals and amounts within 15 percent of the median: payroll, social contributions, rent, loans, leasing, insurance, subscriptions, VAT. Next date = last date + median interval. Card spending is forecast as a weekly average of the last 12 weeks, minus large one-offs you know of. A series that has missed its expected date by more than half its interval has stopped: do not project it, ask. One-off items are excluded. With less than 3 months of history, say the patterns are unreliable and ask for monthly fixed costs.

3. Inflows from open invoices, client by client. list_client_invoices (filter_status: ["unpaid"], per_page: "25"), list_clients for names. For each client, the historical delay = payment date minus due date over paid invoices (payment date from the invoice when available, otherwise the matching credit in transactions). Expected date = due date + this client's median delay (no history: + 15 days). Probability: not yet due 0.95, 1 to 30 days late 0.85, 31 to 60 days 0.70, 61 to 90 days 0.50; over 90 days goes to a "doubtful" line outside the base case. Adjust +0.05 for a client who always paid, -0.15 for a client who was more than 60 days late twice. When the business does not invoice from Qonto, inflows come only from the series learned in step 2 and the answer to step 6.

4. Pipeline. list_quotes (filter_status: ["approved"]) not yet invoiced: ask the expected invoicing date if unclear, then apply the client's payment pattern. Pending quotes only feed the upside scenario.

5. Outflows. Supplier invoices still to pay: list_supplier_invoices has no "unpaid" status, so query to_review, to_approve, awaiting_payment, pending and scheduled separately (per_page: "25"). For each status, follow meta.next_page until it is null; deduplicate invoices by ID. Never skip an invoice solely because matched_transactions is non-empty: a linked payment does not prove full settlement. Deduct only amounts verified as settled through the declared read tools, keeping any remaining debt in outflows. If settlement or the remaining amount cannot be verified, include the full invoice amount, flag the uncertainty and ask the user. Place remaining outflows at their due date or at the user's usual payment habit learned from history, plus the recurring series of step 2.

6. One question for what history cannot know. Hires or departures, a large purchase, a loan or grant expected, an annual bonus, a seasonal dip. If the user wants the fast version, skip and list the assumption.