We’re excited to announce the trial launch of our V&A MCP Server, a new way for AI applications including services such as ChatGPT and Claude to access our collections data more reliably.
MCP (Model Context Protocol) is a lightweight service that provides a standard, consistent way for AI applications to connect to external systems such as the V&A Collections API (Application Programming Interface).
Like many organisations in the cultural heritage sector, we have been exploring how AI applications can offer a new way for users to discover our collections. The challenge is that AI applications are not naturally good at interacting with APIs they’ve never seen before. This often leads to an issue called ‘hallucination’, where the AI application generates incorrect or false results and presents it as factual.
To address this, we’re trialling our own MCP service so that users can simply tell their AI application to query the V&A collections even if it has no prior knowledge of our collections API. We expect this to reduce hallucinations and improve the quality of the results.
Here’s a screenshot of calling the V&A MCP Server Tools to locate relevant objects (and you can see the chain of queries and results):
The V&A MCP server is now available for trial use here: https://mcp.vam.ac.uk/mcp/
You can use this service with any AI application that supports remote MCP server connections by simply adding our server URL to the application’s MCP settings.
As this is a trial service, we’d especially welcome your feedback to tell us if this service is useful, if it is still returning hallucinated results and to help us refine and improve it.
Feedback details and further technical background for those interested in more details are available at: V&A AI Sandpit Updates
