Dataverse Harvard

What Dataverse Harvard MCP server does

Dataverse Harvard provides MCP access to the Harvard Dataverse research dataset repository, which hosts approximately 150,000 datasets. The integration is read-only and uses the Harvard Dataverse API at https://dataverse.harvard.edu/api/ through the Pipeworx gateway.

Its dedicated tools cover the main discovery and inspection tasks:

• Search datasets, files, or dataverses with optional sorting, pagination, page size, and starting offset.
• Retrieve full metadata for a dataset using its DOI-style persistent identifier.
• List the files associated with a dataset.
• Retrieve metadata for a dataverse collection by identifier.

The integration is independent and unofficial; it is not affiliated with, endorsed by, or published by Harvard Dataverse.

How it works

The hosted MCP endpoint is:

https://gateway.pipeworx.io/dataverse-harvard/mcp

An MCP client can connect to that URL directly. The endpoint lists the Dataverse tools along with shared Pipeworx gateway tools such as ask_pipeworx, discover_tools, search_within, and remember/recall. The exact scope is reported by the connection's initialize response. The shared ask_pipeworx tool can route plain-language questions across the wider Pipeworx catalog, including sources beyond Harvard Dataverse.

The full gateway is also available at https://gateway.pipeworx.io/mcp. Both endpoints use the same gateway and data sources; the pack endpoint lists the Harvard Dataverse tools directly, while the full gateway lists tools from all packs directly.

For clients without MCP support, individual tools can be called over HTTP. For example, the search tool is available at POST https://gateway.pipeworx.io/v1/tools/dataverse_harvard_search with a JSON body such as {"query":"climate change"}. Tool details can be inspected through the corresponding gateway HTTP endpoints.

Setup and configuration

A hosted MCP connection can be added with the following URL:

{
  "mcpServers": {
    "dataverse-harvard": {
      "url": "https://gateway.pipeworx.io/dataverse-harvard/mcp"
    }
  }
}

The npm package @pipeworx/mcp-dataverse-harvard provides a local stdio bridge. It forwards MCP calls to the hosted gateway and requires network access; it does not provide an offline mode.

{
  "mcpServers": {
    "dataverse-harvard": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-dataverse-harvard"]
    }
  }
}

Anonymous use is available with a daily limit of 50 calls. A free Pipeworx account increases the limit to 200 calls per day. Add PIPEWORX_API_KEY to the local stdio server's environment to authenticate with a key. Setting PIPEWORX_META_TOOLS=1 makes the shared gateway tools appear alongside the pack-specific tools in the npm bridge.

Tools and capabilities

The dedicated tool surface includes:

• search(query, type?, sort?, per_page?, start?)
• dataset(persistent_id)
• dataset_files(persistent_id)
• dataverse(identifier)

These support repository search, dataset-level inspection, file discovery, and collection metadata lookup. Search pagination and sorting can be controlled through optional arguments. Dataset and file operations use a persistent identifier, while collection metadata uses a dataverse identifier.

Limitations and notes

The integration is read-only and depends on network access to the Pipeworx gateway and Harvard Dataverse API. The local npm bridge has no offline mode. Hosted or authenticated calls are subject to Pipeworx usage limits and account metering. A pack endpoint necessarily includes shared gateway meta-tools, so its advertised tool list is larger than the four dedicated Dataverse tools. For clients using an HTTPS proxy, Node's built-in fetch may require NODE_USE_ENV_PROXY=1 on Node versions that support that setting.