Heads up for judges: the live demo runs on Render’s free plan. If it’s been idle, the first load can take around a minute while the instance wakes up. Please give it a moment; it’s fast after that.
What it does: Taste Match is a simple taste-based recommender. You search for things you like and add them as “taste seeds.” Then you choose what you want recommendations for: brands, destinations, movies, music artists, restaurants, or TV shows. Request recommendations, and Qloo’s Taste AI suggests what you’ll probably enjoy next.
How it’s built: it has two parts. The first is qloo-taste-mcp, a Model Context Protocol server written in Python 3.10+ standard library only (zero third-party dependencies), speaking MCP over stdio with newline-delimited JSON-RPC, so any MCP-compatible agent can call the Qloo tools. It exposes 4 tools: qloo_search (find Qloo entities by name), qloo_recommend (taste recommendations from seed entities), qloo_trending (trending entities in a category), qloo_tags (look up Qloo tags). The second part is a separate web demo in demo/, deployed on Render: it calls the same Qloo client code directly, not through the MCP server.
Testing: 18 unit tests, all HTTP mocked so they run offline. The live demo was smoke-tested with the real Qloo API on 2026-10-29: searching “Taylor Swift” returned 8 real Qloo entities (artist, album, movie, and more, as shown in the search screenshot), and the full flow — search, add one or more seeds, pick a category, request recommendations — returned real recommendations (for example, Blade Runner seeds returned 12 sci-fi movies; Taylor Swift seed with music artists returned 12 artists).
Security: the Qloo API key is stored only in a Render environment variable, not in the repository.
MIT licensed. Built for the Qloo Agentic Hackathon.
Built with: Python, Qloo API, Model Context Protocol, Render.
Built With
- model-context-protocol
- python
- qloo-api
- render
Log in or sign up for Devpost to join the conversation.