We built this demo MCP server as a hands-on way to explore how Model Context Protocol (MCP) servers work in practice.
The server exposes a small, proprietary dataset we maintain internally at Setfive: our lunch ratings. The dataset consists of restaurants we’ve visited as a team, along with a short review from each team member after every visit. It’s opinionated, human-generated, and intentionally imperfect—making it a useful test case for experimenting with structured context, retrieval, and tool-driven interaction.
This demo is not about restaurant recommendations. It’s about understanding how an MCP server can:
Below is a snapshot of the data exposed by the demo MCP server. The dataset is intentionally simple and human scale, made up of restaurants we have visited as a team along with individual ratings and short written notes from each visit. The tables show how structured records like restaurants, locations, and timestamps connect to personal feedback over time, making the data an easy, approachable sandbox for exploring MCP concepts such as context exposure, querying, and tool driven interaction.
In this short demo, we skip the setup and configuration steps and jump straight to the fun part. You’ll see how ChatGPT can natively interact with our MCP server and explore the lunch ratings dataset in real time, asking natural questions and working with the data as if it were built right in.
