Uncovering the Hidden API of Alphaxiv: Our Journey to Create a Python Client

Discover how we explored the hidden API of alphaxiv.org and developed a Python package for seamless access.

3 min readTechnology

Alphaxiv.org offers an API, but accessing it was a challenge. Until recently, the public documentation was lacking, leaving developers to rely on browser DevTools and live traffic analysis to understand its functionality. We identified a crucial endpoint at api.alphaxiv.org/assistant/v2/chat, which utilizes a Server-Sent Events (SSE) stream in response to POST requests, powered by the aurelle-1 model. This exploration led to the creation of aurelle-py, a Python package designed for programmatic interaction with the AI assistant at alphaxiv. Users can inquire about arXiv articles, stream responses, and integrate the tool into their research workflows. While we are not the first to tackle this, we aimed to deliver a well-structured and thoroughly documented solution. The package features synchronous and asynchronous clients, an SSE parser with independent unit tests, Pydantic v2 for data validation, and custom exceptions like AuthError and RateLimitError. Additionally, it includes an MCP server for integration with Claude Desktop and Claude Code. The article details our discovery process, request and response formats, limitations we encountered, and the internal workings of the package. Installation is straightforward: pip install aurelle-py.

Technology