Yale Library is testing an AI-powered connection for deeper searching of library resources

Yale Library is testing an AI integration that could transform how students and researchers search for and surface books and other resources in the library’s vast catalog. The integration uses the model context protocol (MCP), a universal open standard for connecting AI helpers like Chat GPT or Claude to any data source—including, in this case, the highly curated bibliographic data of the library’s catalog. 

“To put it simply, it will allow your AI helper to provide you with responses that are deeply informed by the complex data in Yale Library’s catalog,” said Michael Appleby, Yale Library’s director of software engineering and the tool’s principal developer. 

Students testing the integration reported that it reduced their concerns about the reliability of AI-generated outputs. Being able to see that results were coming from the library’s catalog or digital collections increased their confidence in the quality of the results.  

Currently, a student using an AI assistant to explore a research topic must then go to the library’s website and use Quicksearch, a discovery tool based on keyword matching, to find relevant library resources. The MCP integration brings the catalog search into the AI platform, allowing the student to surface resources with a query like, “What does Yale Library have on this topic?” The responses are shaped and refined by AI’s semantic searching capabilities, which take into consideration context and user intent. 

The library is piloting the MCP integration using Claude, Anthropic’s AI assistant. If it is rolled out more broadly at Yale, students would be able to make the connection to the library’s catalog through their preferred AI platform—including, potentially, Yale’s Clarity platform.  

AI and “The Divine Comedy” 

To demonstrate the integration’s potential, Appleby logged into Claude and queried, “What does Yale Library have on Dante’s Divine Comedy?”   

In response, Claude spun up lists of library resources organized by type, including biographical materials, academic works, reference materials, and early printed editions from the library’s Digital Collections.   

Claude’s categories of results, Appleby noted, “can be very helpful if you are exploring a subject where you’re not a specialist. If you ask for more information about a category, it will run more searches, find more specialized resources, and give you a summary of those results. It encourages you to keep iterating in a way that I think is particularly helpful when you are starting  to plan a research project.” 

Appleby’s searches showed the potential to delve deeply into catalog records to identify resources that would not surface based on keywords alone. His Dante queries, for example, produced foreign language resources, connections based on metadata, and other results informed by the AI model.  When Appleby asked for Yale Library’s oldest commentaries on “The Divine Comedy,” Claude launched six searches, including searches on the names of specific early commentators.  “As a non-expert, I didn’t know the commentators’ names, so I would not have found those resources with a keyword search,” Appleby said. 

Testing the concept 

Chelsea Fitzgerald, the library’s program director for user research and strategy, is testing the integration with library staff and students. Her work has included a week-long “diary study” with ten Yale students who used the integration daily to explore the library’s catalog and digital collections while recording their experiences and reactions. Severalstudents reported that knowing the AI application was drawing from the library’s highly curated data reduced their concerns about the reliability of AI outputs and sourcing.  

Students also found value in being able to go beyond keyword search to find relevant resources; the more streamlined discovery process could be a more effective use of researchers’ time. 

“The hardest part of research is often the beginning: knowing what you’re asking and where to start,” said Lauren DiMonte, associate university librarian for Research and Learning. “Our catalog is enormous, and this tool helps students find their footing faster, so they can move from orientation to inquiry, following their curiosity, making unexpected connections, and developing ideas that are truly their own.”   

What’s next? 

The MCP catalog integration, for which additional testing is planned, is just one of the approaches to AI use in research that the library is pursuing. For example, a year-long trial of Consensus AI, an AI-powered discovery tool designed to accelerate the process of online research, is underway. Chat features, like the Cushing/Whitney Medical Library’s “Ask Harvey” virtual assistant are being developed. And the library continues to develop and test AI applications to research in special collections. 

A critical initiative involves exploring the ability of AI to direct students to the library’s subject specialists and other library experts. “We are thinking carefully and purposefully about how to build the bridges for students and other researchers between AI helpers and the library’s extraordinary resources, both digital and human,” DiMonte said. 

Yale Library supports the ethical and effective use of AI to improve research, teaching, and learning. The library offers guidance, resources, and AI-powered tools to broaden access to collections, facilitate work with data, and unlock new research possibilities. Library staff members help researchers locate and access resources, build AI fluencies, and evaluate the quality and reliability of AI systems. 

Learn more about AI resources at Yale Library. Learn more about AI at Yale. 

—Patricia M. Carey