FinReviewLens

Mobile app review intelligence for a financial services company. FinReviewLens ingests iOS and Android app-store reviews and turns them into an analytics dashboard and an MCP server — so the same review intelligence is available in a browser and inside AI assistants like ChatGPT and Claude.

What's Inside

This is an npm-workspaces monorepo with three apps sharing one analytics core (lib/):

Workspace	What it does
frontend/	React + Vite dashboard with six views (Overview, Review Inbox, Themes, Platform Comparison, Prioritization, Insights)
backend/	Express REST API serving reviews and computed analytics on port 4000
mcp-server/	Model Context Protocol server exposing the same analytics as 13 tools + interactive HTML widgets

The dataset is generated mock data (168 reviews across iOS and Android) seeded by scripts/generate-reviews.js — no real customer data.

Screenshots

Dashboard

Review Inbox	Platform Comparison

Themes	Prioritization

Insights Generator

MCP Widgets in ChatGPT

The MCP server renders interactive widgets (rating summary, top themes, platform comparison, prioritization) directly inside ChatGPT:

Features

• Overview — total volume, average ratings per platform, rating distribution, review volume over time, top complaint and positive themes
• Review Inbox — browse, search, and triage reviews with filters for platform, rating, sentiment, theme, app version, segment, status, and date range
• Themes — reviews clustered by product area (notifications, transfers, biometric login, …) with representative quotes and suggested actions
• Platform Comparison — iOS vs Android complaints, strengths, and theme deltas
• Prioritization — impact/severity matrix and ranked P1–P20 priorities with recommended next steps
• Insights Generator — deterministic executive summaries computed locally from the dataset
• Workflow — update review status (new → reviewed → planned → in progress → shipped/declined) and priority from the UI or via MCP tools

Getting Started

Prerequisites

• Node.js 18+

Run the Dashboard

npm run install:all   # install root + all workspaces
npm run dev           # backend on :4000, frontend on :5173

Open http://localhost:5173.

Run the MCP Server

npm run dev:mcp                    # stdio transport (Claude Desktop, etc.)
npm run dev:http -w mcp-server     # HTTP transport (ChatGPT developer mode)
npm run inspect -w mcp-server      # debug with MCP Inspector

MCP tools exposed: list_reviews, get_review, search_reviews, filter_reviews, get_rating_summary, get_platform_comparison, get_top_themes, get_theme_details, get_low_rating_reviews, get_prioritization, update_review_status, update_review_priority, update_review_theme.

Regenerate the Dataset

npm run seed

Project Structure

├── backend/          # Express API (port 4000)
├── frontend/         # React + Vite dashboard (port 5173, proxies /api → 4000)
├── mcp-server/       # MCP server (stdio + HTTP transports)
│   └── widgets/      # Interactive HTML widgets rendered in MCP clients
├── lib/              # Shared analytics core (store, analytics, insights)
├── data/             # Generated mock review dataset
└── scripts/          # Dataset generator

Tech Stack

React 18 · Vite · Recharts · Express · @modelcontextprotocol/sdk · Zod

License

MIT