MerchantOS — WebMCP Storefront Co-pilot

• 

  AI-generated promo copy and tool test results

• 

  Dashboard with 3 WebMCP tools ready

Inspiration

Small merchants often have overstocked products and no time to write promos. We wanted a dashboard where an AI agent and a human work side by side: the agent proposes actions, the human stays in control.

What it does

MerchantOS registers three WebMCP tools in the browser:

• search_inventory (read-only): filter products by keyword and minimum stock
• apply_bulk_discount (write): discount high-stock, low-sales products
• generate_promo_copy (write): draft short promo copy via an AI endpoint

How we built it

• Frontend: vanilla JavaScript + WebMCP, hosted on Netlify
• AI bridge: Netlify Function /api/promo calls Fireworks AI (DeepSeek v4 flash) with strict limits (max 3 products, max_tokens 200)
• Fallback: deterministic mock copy when AI is unavailable

Challenges

• Keeping AI costs near zero: single endpoint, tiny context, capped output
• Making agent actions visible: every write triggers a UI update so the human can review
• Parsing AI output reliably: strict prompt + lenient parser + safe defaults

What we learned

WebMCP makes agent actions transparent and auditable on a real UI — a practical pattern for human + agent collaboration.

Built With

• fireworks-ai
• javascript
• netlify
• netlify-functions
• webmcp