Most agentic AI demos today focus on synthetic benchmarks, toy calculators, or coding assistants editing small functions.
What does consumer AI automation look like when plugged into real production systems?
I built silpo-agent-cli, a terminal CLI and Claude Code agent skill that interfaces with the official Model Context Protocol (MCP) server of Silpo, one of Ukraine's largest supermarket networks.
The Problem It Solves
Every week, grocery shopping involves 80% repeat purchases: coffee, milk, bread, produce, water. Re-adding these items in a web or mobile GUI wastes 15–20 minutes of repetitive tapping.
By exposing grocery basket operations to an agent via MCP, we can automate:
- Re-aggregating regular cadence purchases.
- Checking for active sales and discounts on staple goods.
- Batch updating the real remote cart.
Safety & Guardrails By Design
Automating purchases must balance speed with financial safety. silpo-agent-cli enforces a strict guardrail:
The CLI stops at "Cart is Ready".
It fills your authentic shopping basket, verifies stock, and prints the summary. The final checkout step, delivery address selection, and payment card confirmation remain strictly in the human user's hands.
Quick Installation
pip install silpo-agent-cli
# or via uvx without installation:
uvx silpo-agent-cli reorder
AI Agent Skill Integration
Point Claude Code at the bundled skill definition:
"Reorder my usual groceries, replace oat milk with brand on sale if available, keep budget under 1500 UAH."
Claude inspects history, balances costs, and fills your basket.
- GitHub Repository — github.com/MIt9/silpo-agent-cli
-
PyPI Package —
silpo-agent-cli - Video Walkthrough — YouTube
Top comments (0)