Not ChatGPT, which makes up restaurants that do not exist. A real AI that reads every Google Maps review and tells you which restaurant is actually good.
On Google Maps every restaurant looks the same. One has 4.3 stars from 500 reviews. The next has 4.5 stars from 4,000. Which one is actually better? A star average mixes the food, the decor, the service and the delivery time into one number, so it cannot answer that question. The answer sits in the review text, and nobody has time to read three thousand reviews.
You type a craving in plain language, in English or in Arabic. Halulu finds the real restaurants near you, reads their Google Maps reviews in both languages, and ranks them by what reviewers actually said about the dish you asked for. Every result carries a Halulu Score, the Google rating beside it for contrast, the distance, and two verbatim review quotes. One quote is positive and one is negative. The quotes are the proof.
The Halulu Score rates one dish at one restaurant. It does not rate the restaurant as a whole. Hashem Restaurant in downtown Amman shows the difference. Google gives it 4.4 stars across 3,448 reviews. Halulu scores its falafel 4.7. Reviewers praise the food almost unanimously, and the lower star average reflects plain decor and plastic sheets on the tables. If you came for the falafel, the star average pointed you the wrong way.
Halulu runs a Model Context Protocol server, so an AI assistant can search restaurants, save favorites and share a result for you without leaving the chat. Add the connector at https://www.halulu.food/api/mcp , or read the step by step guide on the connect page.
Halulu speaks English and Arabic, with full right to left support. It works anywhere Google Maps carries restaurant data. The first markets are the Gulf and the wider Middle East.
The whole integration surface is documented. Start at the developer portal, or read the agent integration guide, which lists every tool. The HTTP surface is described in OpenAPI 3.1, authentication is walked through step by step in auth.md, and the API catalog indexes every one of these documents. There is no REST API and no SDK: the MCP server is the whole of it.