Duka Stock MCP Server
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Duka Stock MCP ServerSold 3 milk and 2 bread"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Duka — Stock, spoken.
Watch the 2:06 demo · Judge testing guide · MCP integration · MIT license
A working voice-and-text stock assistant for a small shop. Record sales and deliveries, check low stock, and prepare reorder drafts. Every stock change requires explicit review. Products, suppliers, and prices are fictional sample data for a Lusaka corner shop.
Built as an Alexa+ web simulation for the Build, Ship, Shape: Amazon Developer Hackathon. This version uses browser speech recognition, a deterministic command parser, and a real HTTP MCP tool endpoint. It is not connected to Alexa, an Echo device, an LLM, or a supplier.
Try it
Type
Sold 3 milk and 2 bread.Check the before-and-after quantities, then choose Confirm update.
Type
Received twelve breadand confirm.Ask
What is running low?, then choose Reorder draft.Review and save the draft. Open Reorders to download its CSV.
Open Activity to see confirmed changes. Reloading preserves the shop.
The microphone writes a transcript into the input. Check it before sending. Browser support varies, so typed commands and manual product updates are also available. Voice recognition may send audio to the browser vendor's speech service.
Related MCP server: Inventory MCP Server
Run locally
Requires Node.js 22.13 or later and npm.
npm ci
npm run build
node --import ./scripts/sites-env.mjs ./node_modules/wrangler/bin/wrangler.js d1 execute DB --local --persist-to .wrangler/state --config dist/server/wrangler.json --file drizzle/0000_nervous_black_queen.sql
npm run devOpen http://localhost:5173. Run the initial SQL once for a fresh local database. The Sites development server supplies a sample signed-in identity. No model API key is needed.
Sites handles hosted builds and generated D1 migrations. The logical binding in .openai/hosting.json is DB; do not replace it with a Cloudflare database ID. Preserve the supplied Vite plugin and build infrastructure.
Project map
Location | Purpose |
| Inventory, voice input, review, drafts, activity |
| Catalog, command grammar, quantities, reorder rules |
| Identity, persistence, atomic confirmation |
| MCP initialization and four tools |
| Schema and generated migration |
| Domain tests and isolated Worker integration checks |
| Architecture, protocol usage, validation |
| Devpost draft, video script, friction log, checklist |
Verify
npm test
npm run typecheck
npm run build
npm run test:integration:isolatedFor HTTP/database checks, follow the isolated test setup. They change sample stock and must not target real data or the usual development preview.
Current boundaries
English stock commands, whole units, ten sample products. Catalog editing, offline use, roles, and supplier integrations remain future work.
Reviews expire after ten minutes. Conflicting updates require a fresh review. Replaying a successful confirmation cannot deduct stock twice.
Orders are unsent drafts; no payments or supplier messages are made.
Hosted access uses Sites account authentication. Tool responses provide a refresh-safe review link. An external Alexa client still needs supported authentication.
The public English demo shows typed input with fictional shop data and synthetic narration. Merchant trials and microphone/device compatibility studies remain future work.
Final Devpost submission awaits the rules-and-terms agreement.
Project code is MIT licensed. Bundled dependencies retain their licenses; see third-party notices.
This server cannot be deployed
Maintenance
Related MCP Connectors
Approval-first store operator MCP: drafts, reads, and shop tools. You still approve.
- TightlyOAuthio.tightly
Read Tightly stock, sales, suppliers and purchase orders, ask Tightly, and act once you confirm.
AI inventory management: stock, Demand ESP reorder recommendations and purchase orders.
- TecoposOAuthcom.tecopos
Run your Tecopos business from your AI: sales, inventory, cash, customers, purchases and reports.
Related MCP Servers
- FlicenseNot gradedqualityBmaintenanceEnables AI-powered warehouse inventory management through MCP, with tools to search products, check stock levels, receive and issue stock, and add products under human-in-the-loop confirmation.-
- FlicenseNot gradedqualityCmaintenanceEnables querying and administering a small business inventory through SQLite-backed tools, including stock levels, low-stock alerts, restock recommendations, product movements, and controlled write operations.-
- AlicenseAqualityBmaintenanceEnables AI agents to manage inventory conversationally: list and look up products, check current stock levels, record inbound and outbound movements idempotently, review movement history, and surface items below their minimum threshold. Every figure is grounded in tool calls rather than model estimation, and invalid operations return explicit error codes the agent can reason about.5MIT
- AlicenseNot gradedqualityBmaintenanceEnables users to query and manage warehouse inventory through MCP by searching SKUs, viewing item and stock details, identifying low-stock items, summarizing warehouses, and creating reorder requests.MIT