Open Museum MCP

What cfpramod/open-museum-mcp does

cfpramod/open-museum-mcp combines artwork searches from registered cultural collections behind one MCP interface. Its live sources include The Metropolitan Museum of Art, Rijksmuseum, the Smithsonian, Cleveland Museum of Art, the Art Institute of Chicago, SMK, the Walters Art Museum, Wellcome Collection, the National Gallery of Art in Washington, Harvard Art Museums, the J. Paul Getty Museum, the Yale Center for British Art, Wikimedia Commons, and Europeana.

Results use a shared structure rather than each institution's native response format. Records can include the title, artist and attribution details, dates, medium, region, period, image URLs, source links, and verified license information. The normalized data is suited to research workflows that compare works across institutions or need consistent metadata for generated references.

How it works

The server queries registered collection sources and filters results through a per-museum rights gate. Only records that meet the configured open-access rules are returned; missing, ambiguous, or unverified rights indicators are rejected instead of being treated as open by default. The rights result includes verification data such as the license type, source field, timestamp, and confidence when available.

Artwork dates can be represented as ranges and include dynasty-aware parsing for periods such as Tang, Edo, Safavid, and Mughal. Searches can also normalize regional terminology and distinguish named artists from anonymous, workshop, attributed, and “after” works. A date range can be applied with inclusive minimum and maximum years, including negative integers for BCE dates.

Setup and configuration

Run the published package through an MCP client's standard server configuration:

{
  "mcpServers": {
    "open-museum": {
      "command": "npx",
      "args": ["-y", "open-museum-mcp"]
    }
  }
}

The package is fetched by npx when the server starts. Claude Desktop also has an installable .mcpb bundle from the project's latest release, while VS Code and Cursor provide install links that prefill their MCP setup. For a local checkout, clone the repository, run npm install, build it with npm run build, and point the client at dist/server.js using Node.js.

Some museum APIs are available without credentials, while others use an opt-in free API key. Configure any required credentials according to the relevant source setup before querying those collections.

Tools and capabilities

• search_artworks searches registered museums and sources with optional museum, image, result-limit, and inclusive year-range filters. It returns only records that pass the rights gate.
• get_artwork retrieves one artwork using its normalized identifier.
• cite creates deterministic full, caption, or short citations from an artwork record.
• discover_random supports discovery by region and period, including workflows such as finding an Edo-period counterpoint to a European work.
• museum://{code}/{id} resources provide bookmarkable artwork identifiers for use during a session.

These capabilities support cross-institution searches, comparisons by period or region, rights-cleared image selection, and references formatted for papers, slide captions, or inline use.

Limitations and notes

The rights filter is intentionally conservative: a potentially useful record can be omitted when its open-access status is absent or unclear. Coverage depends on the registered sources and their APIs, and some sources may require a free API key. The server is designed for records that pass its open-access rules; it is not a general search interface for every item held by the listed institutions.