Connect your own AI assistant —Claude or ChatGPT— to Gastos Compartidos and manage your expenses by talking: log them, read a receipt from a photo, ask for the month's summary. With scoped permissions and full traceability.
Connect your own assistant —Claude or ChatGPT— once, with your permissions, and operate from the chat you already use. The app is there when you want the month's detail in screens and charts.
Connect your account to ChatGPT (developer mode) via MCP server: it logs expenses, checks balances, and builds reports on your behalf, always with your authorization.
Claude (Desktop, web, or Code) and any MCP-compatible client —Cursor or your own agents— operate the app with your scopes.
Log by talking, read receipts and screenshots, import from a statement, summarize, and flag — all in natural language, no forms. Every sensitive action asks for confirmation before touching the sheet.
Tell it the amount; the category comes from context and the split applies per your agreement. Confirm with one tap.
You"Paid $94,800 for winter clothes"
Agent"Logged under Clothing · 50/50. Confirm?"
Snap a photo of the receipt. Your assistant extracts amount, date, merchant, and category. A Plus and Profesional plan feature.
You📎 receipt-supermarket.jpg
Agent"$48,350 in Groceries. May 15. Is this the sheet?"
Hand it a screenshot of your Mercado Pago or wallet activity and tell it which to log. It detects amount, date, and merchant for each; you review and confirm.
You📎 mercadopago-screenshot.png
"Log these three to the home sheet"
Agent"Found 3 movements: $12,400, $8,900, and $23,100. Log them 50/50?"
Attach your bank or card statement and filter what you want: "the USD charges, to the 2026 USD expense sheet". It builds the list and waits for your OK.
You📎 statement-june.pdf
"Log only the USD charges to the USD 2026 sheet"
Agent"Found 5 USD charges (US$ 214 total). Logging them to USD 2026 — confirm?"
Ask for the cycle summary, compared to last month, by category or by participant.
You"How are we doing this month?"
Agent"You're at $739,450 in May, 12% less than April."
If an observation gets stuck, your assistant summarizes the context and drafts a clear reply that you edit before sending.
You"Help me reply to the health observation"
Agent"I'll put together the summary and a draft for you to review."
Every interaction with your assistant respects three principles. No exceptions.
Your assistant only accesses the sheets you authorize and only the actions you enable. Each connection (ChatGPT, Claude, or any MCP client) has its own revocable token.
Sensitive actions (logging an expense, opening an observation, accepting a proposal) require your confirmation in the chat before touching the sheet.
Every assistant action is recorded in the history with a clear mark of who ran it (you, your assistant, the other party). Your data is not used to train models.
We're validating answer quality, conversational coverage, and OCR accuracy. When it leaves Beta, the AI assistant and receipt OCR become part of the Plus and Profesional plans. The Free plan keeps the rest of the features.
It's free. No card. The numbers are always computed by the app, not the AI.