what is auto fr?

This project, called FinGate MCP, is a hackathon entry built for a banking and fintech compliance track tied to the EU AI Act, a European law that sets rules for high-risk AI systems such as those used in credit scoring. The idea is that AI agents connecting to banking tools through the Model Context Protocol normally have no built-in oversight, and this project adds a control layer in between so risky actions get checked before they happen. The core feature is a human-in-the-loop step: if an AI agent tries to do something risky, like transferring more than 10,000 dollars or acting for a flagged user, the system pauses the request and asks a person to approve or deny it through a separate route before continuing. It also includes a tuning system that adjusts its own risk thresholds over time based on past approvals and denials, aiming to reduce false alarms while staying cautious. To stop a human approval from being tampered with, the exact parameters shown to the approver are matched to the ones sent to the banking system using a cryptographic hash. Every intervention is also logged to a local database for audit purposes. The codebase is a Next.js application paired with a separate MCP server built on a framework called NitroStack, using Prisma for the database and Zod for validating tool inputs. Getting it running requires Node.js 20 or newer, the Bun runtime, and the Prisma command line tool, then starting the MCP server and the main app in separate terminals. The README also documents how the project divides work across four hackathon team members covering backend protocol work, orchestration logic, frontend widgets, and compliance testing, along with a four-phase build plan and a script for running adversarial tests against the approval gates.