cursor-byok

Why we included this project

Cursor users who want to stop being locked into the editor's bundled model subscriptions can run this local gateway instead. It sits on your machine, takes requests from Cursor, and forwards them to the OpenAI- or Anthropic-compatible endpoints you configure, so the editor works with your own API keys, custom endpoints, and model IDs while agent features like tool calling, Skills, and MCP keep working. If you already run LiteLLM, vLLM, or Ollama behind a proxy, this is a clean way to reuse those backends inside Cursor without exposing localhost to the public internet. The dashboard also helps you manage several model configurations, batch-test them, and benchmark time to first token and generation speed, which is handy when you're comparing providers before committing to one.