Expense Organizer is a shared-expense platform developed and published by Rusmex Software Solutions, and our production reference implementation for application-native AI: Claude operates through the same domain model, permissions, and workflows used by the application itself.
Production use case: divorced parents reconciling shared child-related expenses across households. The FemExpensor mobile app captures receipts; the web application keeps the ledger; Claude and other assistants reach both through MCP.
Every capability is defined once, in application code, and published as a tool. Seven server tools reach the outside world over MCP, behind OAuth 2.1 with PKCE. Inside the app, each screen publishes its own tools only while it is on screen: an open expense form offers fill and populate, the grid offers filter, sort and status changes, the drafts queue offers promote and merge. The in-app agent’s tool list is assembled fresh for every message from what is actually mounted, and the server tools are projected into it from the same code, with one schema and one description.
AI is a supported client of the application, not a bypass around it. Every call is authenticated, scoped to the signed-in member’s family group, and subject to the same business rules and event log as the web UI. Money math, splits, statuses and permissions stay in application code; the model handles capture, reading, summary and conversation — and when it cannot tell which answer is right, it records nothing rather than guessing.
Blazor and .NET web application, SQL Server, native mobile apps. Model-aware but not model-locked: Claude is used where it is part of the implementation; the MCP and tool layer works with other compatible assistants.
The platform is configurable and can be branded for a business. The infrastructure — domain model, permissions, audit, MCP — is shared across configurations. The platform customized for Divorced Families is live at family.expenseorganizer.com. Events (group weekend outings, weddings) and Rental Properties are future customizations of the same core platform.
Forty-five minutes. We look at the application you have and tell you where AI belongs in it, what it needs, and what should stay deterministic.