Test our new AI integration

Published:

9/1/26

Imagine asking an AI assistant to summarise travel diaries held in the archives. Or asking it to find out who sat on the board of Swedish joint-stock companies in the 1910s, or to read and summarise court records from the 1600s. This and much more is now possible with the National Archives' new MCP server – a connection that lets AI assistants such as ChatGPT, Claude and Vibe access sources in the National Archives' collections directly.

We invite you, as an archival researcher, to test this feature and tell us what works, what doesn't, and what would make the service even more useful for your research.

What is an MCP server?

MCP (Model Context Protocol) is an open standard that lets an AI assistant connect to external tools and data sources – rather than only answering from what it learned during training. With an MCP connection to the National Archives, your existing AI assistant can actively search our finding aids and collections, retrieve images, and read handwritten documents that have been transcribed with handwritten text recognition, all while you chat.

If you already use ChatGPT, Claude or Vibe in your research, the step to connecting the National Archives' MCP server is a small one.

What can you do with the National Archives’ MCP server?

During the test period, the server gives your AI assistant access to:

• Finding aids and metadata for the 250,000 archives in the National Archival Database (NAD).
• Millions of handwritten documents that have been transcribed using handwritten text recognition.
• Over 73 million digital images of archival records.
• Several of the National Archives' comprehensive archival guides.
• Over 30 registers and datasets drawn from the archives, including the Dictionary of Swedish National Biography, medieval charters and parchment wrappers, the Swedish topographical register, church records, court records, seamen's registers, field ranger rolls, and suffrage petition lists.

See the full list on Huggingface

Who is it for?

The beta test is aimed primarily at researchers and graduate-level students - doctoral candidates, master's students and established researchers. If you carry out systematic archival research without an academic affiliation – for example as a journalist, investigator, or genealogy or local-history researcher – you're welcome to test it too.

How to get started

1. Choose your AI assistant. The MCP server can be connected to the Pro versions of Claude, as well as to ChatGPT and Vibe (formerly Le Chat).
2. Connect your AI assistant to the National Archives' MCP server by following our quick reference guides (see below).
3. Test it! Use the MCP server in your actual research over a period of time. The more real research questions you bring to it, the more valuable your feedback will be.
4. Share your experience. After you've used the service for a while, we ask you to complete a short survey. The survey is in English and takes about 10 minutes to complete.

We will also be inviting participants to interviews and a workshop, for those who want to share their feedback in more depth.

What does "beta" mean?

The National Archives' MCP server is experimental software at an early stage of development. Performance is still limited, and you will likely encounter both bugs and limitations. It's valuable for us to let real users test a product already at the beta stage – that is, before it's fully finished. This lets us gather feedback early and adapt the service to users' needs.

Please keep in mind that AI assistants can misunderstand, misquote, or simply be wrong.

Timeline

The beta test begins in September 2026 and runs through the end of the year. The survey will be available throughout the entire period. If you choose to leave your contact details when completing the survey, we may get in touch with you for a follow-up interview or workshop later on.

Questions?

Get in touch at ai@riksarkivet.se if you have questions about the beta test.