MCP (Model Context Protocol) server for Dataverse

Credits: this work is initially funded by the SSHOC-NL project developing Semantic Croissant. The first version of Croissant export for Dataverse was implemented by Philip Durbin (Harvard IQSS) and Slava Tykhonov, CODATA AI lead and Harvard Dataverse Ambassador. Starting from August 2025, further development of Dataverse MCP server with variables support by CDIF framework is funded by CODATA.

Croissant is a special language for machines, built on top of Schema.org. With Croissant, we aim to solve multilingual challenges and finally speak the same language across the planet. Even if it’s artificial.

Getting started with mcp.dataverse.org

When getting started, we recommend the public MCP server for Dataverse at . (Below you’ll also find instructions on how to run the MCP server locally.) You can visit https://mcp.dataverse.org/tools for an inventory of available tools.

You will need an MCP client with AI agent support such as Cursor, Visual Studio Code, Claude Desktop, Windsurf Editor, Zed, or Codex CLI.

(Optional) Command line test

Before you get too far into configuring your MCP client, you could try this quick test to get information about a dataset by passing its DOI.

curl -X POST "https://mcp.dataverse.org/tools/get_croissant_record" -H "Content-Type: application/json" -d '{"doi":"doi:10.7910/DVN/WGCRY7"}'

Configuring your MCP Client

You’ll be using https://mcp.dataverse.org/sse as the URL and SSE (Server-Sent Events) as the type of MCP server.

Click the arrow to expand instructions for your MCP client.