
SiteCheck AI
Interface UIpar frankreno
Construction site inspection tracker with interactive AI widgets. Log deficiencies, view dashboards, and generate PDF reports through ChatGPT, Claude, or any MC
What it does
SiteCheck AI is a construction site inspection and deficiency tracking app that demonstrates the MCP Apps (ext-apps) protocol. It runs as both a full Next.js web app and an MCP server, so site managers can log and track deficiencies either through a browser UI or by chatting with an AI assistant. Items created via ChatGPT or Claude appear instantly in the web app — same SQLite database, multiple interfaces.
It ships with 11 MCP tools. In OpenAI/Claude mode, tools render interactive widgets (dashboards, tables, forms, severity pickers, photo uploads) directly inside the chat. In generic mode, all tools return plain-text responses that any MCP client can use.
Key features
- Interactive widgets — ChatGPT and Claude render dashboards, deficiency tables, stat cards, severity pickers, and photo upload forms inline in the conversation
- Deficiency lifecycle — Create, filter, update severity/status, attach photos, and close deficiencies through natural language prompts
- PDF report generation — Ask the AI to generate an inspection report; a download link appears instantly
- Dual-mode MCP server —
/mcpfor any client (plain text),/mcp/openaifor ChatGPT and Claude widgets - Platform adapter pattern — Core tool logic is shared; adapters control registration and response format per client
Installation
Requires Node.js. No external services or API keys needed.
git clone https://github.com/frankreno/ai-sample-app
cd ai-sample-app
npm install
npm run setup # creates SQLite DB and seeds sample projects
npm run dev # web app on http://localhost:3000
npm run mcp # MCP server on http://localhost:8787
Claude Desktop (via stdio proxy using uvx mcp-proxy):
{
"mcpServers": {
"sitecheck-apps": {
"command": "/Users/YOUR_USERNAME/.local/bin/uvx",
"args": ["mcp-proxy", "--transport", "streamablehttp", "http://localhost:8787/mcp/openai"]
}
}
}
ChatGPT: Expose the MCP server publicly (ngrok http 8787), then add the ngrok URL + /mcp/openai in ChatGPT Settings → Apps.
Supported hosts
Claude Desktop (via mcp-proxy stdio bridge), ChatGPT (via ngrok + HTTP), and any MCP client that supports Streamable HTTP.
Installation rapide
npm install && npm run setup && npm run mcpInformations
- Tarification