aaryanptl/openrouter-mcp

This project lets an AI assistant like Claude or ChatGPT answer questions about the more than 300 AI models listed on OpenRouter, a service that gives access to many different AI models through one place. It works using the MCP standard, which lets an AI tool call small external tools mid conversation, and each tool here also comes with a small visual widget built in React so the answer shows up as a card or chart instead of plain text. It is built to be deployed on a platform called Manufact, though it can also run locally. Under the hood, everything relies on one public OpenRouter web address that lists every model along with its price per token, how much text it can handle at once, what kinds of input it accepts, and benchmark scores, and this project fetches that list once every six hours and reuses it, so no login or API key is needed for the main features. The five tools include one that shows a single model's price and capabilities, one that compares two to four models side by side, one that filters for the cheapest model matching certain needs like vision support, an interactive calculator that recomputes cost live as you change token counts, and a ranking of models by a coding benchmark leaderboard. To try it yourself, you install the dependencies with npm and run it locally, which starts a small server along with an inspector tool you can use to call each of the five functions directly with sample inputs. To put it online, you push the code to GitHub and connect the repository through the Manufact platform, and no extra setup is needed unless you later want to add real model calls, in which case an OpenRouter API key would be required. The README notes a few details, like prices being stored as tiny decimal numbers you multiply to get a normal dollar figure, and that not every model has a benchmark score available.