Skip to main content

HUN-REN Data Repository Platform

ARP Dataverse MCP server

ARP Dataverse can currently be used in several ways: primarily through the graphical user interface, from the command line using pyDataverse for more experienced users, or from any programming environment through the API defined in OpenAPI.

The ARP Dataverse MCP server provides an additional access method for users of Codex, Claude Code, OpenCode, and similar agent-running environments (“agent harnesses”). It enables dataset editors and data stewards to automate Dataverse-specific tasks without programming knowledge, simply by interacting with and prompting the agent. Such tasks may include searching for and editing datasets, adding files to datasets, generating statistics about datasets, or publishing datasets.

How does ARP Dataverse MCP server work?

To use the ARP Dataverse MCP server, its endpoint must be configured in the relevant agent environment. For example, with Codex:

codex mcp add my-server --url https://repo.researchdata.hu/mcp

For Claude Code:

claude mcp add --transport http --scope user my-server https://repo.researchdata.hu/mcp

For OpenCode:

opencode mcp add my-server --global --url https://repo.researchdata.hu/mcp

Alternatively, you can simply ask the agent itself:

Install the ARP Dataverse MCP server available at: https://repo.researchdata.hu/mcp

Or, as a last resort, follow the documentation of the relevant agent environment.

ARP also offers another MCP-based solution called “VibeARP MCP.” It can perform tasks similar to those supported by ARP Dataverse MCP, but there are a few differences:

  • VibeARP edits packages in RO-Crate format and, through the AROMA editor, can display dataset metadata and make it directly editable by the user. ARP Dataverse MCP—and Dataverse itself—does not provide this type of integration. In this case, users need to open the Dataverse interface whenever they want to inspect the data. Furthermore, changes made through the Dataverse graphical interface are not immediately visible to the agent; this type of synchronization is left to the user and the agent.
  • VibeARP MCP is installed locally on the user’s machine, whereas ARP Dataverse MCP is a centralized cloud-based MCP service.