Oxford University Press announced on Tuesday a three-month pilot, which began in October and runs through December, letting students and researchers at Harvard Library, the Wharton School of the University of Pennsylvania, and the University of Oxford search licensed OUP content inside ChatGPT Edu, the version of OpenAI’s chatbot sold to universities. Selected users can ask ChatGPT Edu questions and get answers drawn from OUP books and journals their institution already buys or subscribes to, with citations and links to the full text on Oxford Academic, OUP’s online publishing platform. Behind the scenes is a Model Context Protocol (MCP) server OUP built with Silverchair, which hosts its online content; when a user asks a question, OUP’s server searches the press’s content, picks the most relevant sources, and passes them to ChatGPT for its answer. OUP said its own survey found researchers are using AI more often but still distrust it, and pointed to growth in “zero-click” searches. Tanya Laplante, director of product platforms for OUP’s academic division, said the pilot aims to learn what real-world use of publisher MCPs looks like and what benefits it drives. The press and the three institutions will track usage and collect feedback, including whether citations from a known publisher motivate click-throughs and whether access to OUP information improves ChatGPT’s answers. Richard Ovenden, head librarian at the University of Oxford, said the collaboration will help libraries develop approaches ensuring students and researchers