Xchat1/cursor2api-go
Free gemini-3-flash | cursor不倒我不倒🙏 ❌ 这次倒下啦
What it solves
Cursor2API is a Go service that turns Cursor Web into an OpenAI-compatible API endpoint. Instead of using the original Cursor Web interface directly, you can call it with standard OpenAI chat/completions requests, making it easy to plug into any tool or app that already speaks the OpenAI API format.
How it works
The service runs as a local HTTP server (default port 8002) and translates incoming OpenAI-style requests into Cursor Web calls. It supports non-streaming and streaming chat, tool/function calling (tools, tool_choice, tool_calls), API key authentication, and automatic exposure of a -thinking variant for each base model. Configuration is done via environment variables (PORT, API_KEY, MODELS, TIMEOUT, DEBUG, KILO_TOOL_STRICT). It also includes a health check endpoint and a debug mode that logs detailed request info. Deployment is possible via source build, Docker, Docker Compose, or a Linux systemd service.
Who it's for
Developers who want to use Cursor Web's capabilities (currently the Gemini 3 Flash model) inside any application that supports a custom OpenAI API endpoint, such as ChatGPT Next Web, Lobe Chat, or other agent orchestrators. It is also useful for those who need a lightweight, self-hosted gateway with tool-calling support and minimal setup.
Highlights
- OpenAI
chat/completionscompatible API, so existing OpenAI clients work with minimal changes. - Supports streaming and non-streaming responses.
- Full tool-calling support (
tools,tool_choice,tool_calls). - Automatically publishes a
-thinkingmodel (e.g.,gemini-3-flash-thinking) derived from the base model. - Configurable via environment variables, including
KILO_TOOL_STRICTfor better compatibility with orchestrators that force tool use. - Available as Docker image, Docker Compose, systemd service, or local scripts (Linux/macOS and Windows).
- Built-in health check and optional debug logging for easier troubleshooting.
Related
- Project
- Project
- Project
- Project