Main Street MCP

Put any local business on Alexa+ in five minutes, with one JSON file.

Big brands are building Alexa+ experiences. The corner pizzeria, the barbershop and the family clinic can't afford to, so voice assistants send their customers to the chains. Main Street MCP is an open-source, self-hosted MCP server (spec 2025-11-25, Streamable HTTP). It turns a business's menu, hours and booking rules into tools Alexa+ can call.

"Alexa, order a large margherita and two garlic knots from Tony's for pickup." → "That's 1 Margherita Pizza (Large), 2 Garlic Knots for Renu, $31.00, ready at 7:20. Should I place it?" → "Yes." → The order appears on Tony's live order board.

Features

Built for voice:

• Every result includes a short speech line the assistant reads aloud as written.

• Orders and bookings always ask the customer to confirm first, so a misheard request never turns into a real order.

• If a request is ambiguous ("margherita" when there's a large and a small), the server returns a question for the customer instead of guessing.

• Order IDs are short and easy to say ("A-M-X-T"), with no letters that sound alike.

• Tool annotations (readOnlyHint, destructiveHint) tell the assistant which actions are safe.

Works for any business: Restaurants use the menu and orders. Barbers, salons and clinics use appointments. See config/barbershop.example.json. Tools register automatically based on what the config enables.

Related MCP server: VoiceCart

Voice demo (simulated Alexa+)

Open http://localhost:3000/voice, tap the mic (Chrome, Edge or Safari) or type, and talk to Tony's. The right-hand panel shows every MCP tool call the assistant makes. Open /dashboard next to it and orders appear live.

The assistant connects to this server as an MCP client over Streamable HTTP, the same way Alexa+ does. It has two brains:

cp .env.example .env      # add your AWS keys (.env is git-ignored)
npm run start:bedrock

Quick start

npm install
npm start                       # MCP at /mcp · dashboard at /dashboard · voice demo at /voice
npm run smoke                   # end-to-end test using the official MCP client

Use your own business:

BUSINESS_CONFIG=config/barbershop.example.json npm start

Architecture

Voice web app (/voice) ──▶ /api/assistant ──(MCP client)──┐
Alexa+ / any MCP client ──Streamable HTTP──────────────────▶ /mcp  (McpServer, one per session)
                                               │
                                     tools.js (voice-first rules)
                                               │
                              business.js (hours, menu matching, slots)
                                               │
                                store.js ──change events──▶ /api/events (SSE) ──▶ Owner dashboard

Roadmap

• Simulated Alexa+ voice web app (demo brain + Amazon Bedrock Converse tool use)

• Deploy to AWS (AgentCore Runtime or Lambda) with DynamoDB storage

• Google Sheet as the menu source, so owners edit a spreadsheet instead of JSON

• Square/Toast connectors and SMS "order ready" notifications

License

MIT