Dataville

What the Dataville MCP server does

The Dataville MCP server exposes Dataville’s data-source API to MCP-compatible clients. Dataville provides one interface for public datasets including Wikipedia, arXiv, Project Gutenberg, US Census, USDA FoodData, Papers with Code, SEC EDGAR, OpenAlex, PyPI, and Stack Exchange. The MCP package lets an agent choose a source and retrieve records through tool calls rather than requiring a separate connector for each dataset.

Typical requests include finding an arXiv paper, checking a company’s SEC filing, retrieving Census statistics, looking up a PyPI package, or listing books from Project Gutenberg. Responses can include links back to the relevant source, making results easier to verify.

How it works

The Dataville MCP server can run as a local stdio process launched with npx, or clients can connect to Dataville’s hosted MCP endpoint at https://api.dataville.com/mcp. Local execution sends the configured API key as a Bearer token to the Dataville API. The server does not create, update, or delete records.

Dataville supports both targeted searches and structured queries. For a specific item, keyword search is the dependable route. SQL queries are useful for filtering, counting, joining, paging, and retrieving lists from stored tables. The service can also return CSV or Parquet exports and supports SQL through its broader API, although the MCP tool interface described here exposes read-only selection queries.

Setup and configuration

Local use requires Node.js and a Dataville API key obtained from the Dataville dashboard. The package does not need a separate installation when started with npx:

{
  "mcpServers": {
    "dataville": {
      "command": "npx",
      "args": ["-y", "@dataville/dataville-mcp"],
      "env": {
        "DATAVILLE_API_KEY": "dataville_your_key_here"
      }
    }
  }
}

DATAVILLE_API_BASE_URL is optional and defaults to https://api.dataville.com. It can point to a local backend during development. The hosted endpoint accepts anonymous requests with limits; supplying an Authorization header enables the account’s full quota.

The README provides setup instructions for Claude Desktop and Claude Code, plus direct installation links for Cursor and VS Code. After changing a local configuration, restart the client or session so the tools load.

Tools and capabilities

The Dataville MCP server provides four read-only tools:

• list_dataville_sources lists available sources and their queryable tables.
• describe_dataville_source explains a source’s keyword format, example request, and table columns.
• search_dataville searches one source by keywords and can return a summary.
• query_dataville runs read-only SQL SELECT statements over stored tables.

query_dataville returns no more than 1,000 rows per call and is billed according to rows returned. Census and news are fetched live per search rather than covered by stored-table queries.

Limitations and notes

Stored data is not uniformly complete: some sources are bulk-loaded, while others contain records previously fetched. Consequently, a lookup for one particular item should use search_dataville instead of assuming a stored SQL table contains it. Query results are limited to 1,000 rows per call, and row-based billing applies.

The API key is required for the local configuration and is sent only to the configured API base URL. HTTPS is required for remote base URLs; plain HTTP is accepted only for localhost-style addresses. The tools are read-only, so this server is appropriate for research and retrieval, not for changing source data.