Half Price MCP
Let AI agents search weekly supermarket deals.
Under the hood
1/6 Collect
A scraper reads the half-price deals from Coles and Woolworths. It writes them to a JSON file.
How it works
Hybrid search
The search_deals tool takes a text query. It joins vector search and full-text search in one SQL query. A retailer filter and a maximum price are optional.
Embeddings in Postgres
Titan Text Embeddings V2 on Bedrock makes the vectors. Aurora PostgreSQL with pgvector stores them. An HNSW index finds the nearest deals.
Prices only from data
Each deal has a retailer, a price, valid dates, and a source URL. Each result has an as_of date. A model never creates a price.
Prompt injection defense
Text from store websites is data. The ingest step cleans it. Each tool description tells the agent to not follow instructions in it.
Serverless on AWS
AWS Lambda functions run in Sydney (ap-southeast-2). API Gateway exposes the MCP server over HTTPS.
Repeatable ingest
A scraper writes the weekly deals to a JSON file. The ingest function cleans, embeds, and upserts them. A second run adds no duplicates.
Half Price MCP is an MCP server. An AI agent calls it to find half-price deals at Coles and Woolworths. The server turns each deal into an embedding and keeps it in a vector database. With these results, an agent can plan a shopping basket or suggest when to buy an item. I built it to show a full AI system on AWS: data collection, retrieval, and a tool interface.
Connect your agent
The server uses MCP over streamable HTTP. The URL is https://halfprice.terrydo.dev/mcp. No token or account is needed.
Claude Code
Run this command.
Terminal claude mcp add --transport http half-price https://halfprice.terrydo.dev/mcpOther MCP clients
Add this entry to the MCP configuration file of your client. For Claude Code, the file is .mcp.json. Other clients can use a different file name or key names.
.mcp.json { "mcpServers": { "half-price": { "type": "http", "url": "https://halfprice.terrydo.dev/mcp" } }}Test with MCP Inspector
Run the Inspector. Select the Streamable HTTP transport. Enter the URL. Then call the search_deals tool.
Terminal npx @modelcontextprotocol/inspectorUse the tool
The server has one tool: search_deals. It takes a query. The retailer and max_price arguments are optional. Each result has an as_of date. Each deal has a retailer, a price, valid dates, and a source URL. Try this prompt:
Prompt Find coffee deals at half price. Plan a basket for under $40 and tell me which items to buy this week.
Built with