EveryMuseum MCP server

An MCP server that puts the EveryMuseum open-access art archive inside any MCP-capable client (Claude Desktop, Claude Code, Cursor, …). Once added, you can converse with millions of public-domain and openly-licensed artworks from The Met, Rijksmuseum, the Art Institute of Chicago, the V&A, Getty, Smithsonian, Harvard, Cleveland, and more:

• Pull up images — search by artist, culture, period, medium, or theme and see real artwork images inline in the conversation.
• Learn — get full curatorial detail (description, medium, dimensions, culture, license, source link) and a high-resolution image for any work.
• Go deeper — find visually and thematically similar works.
• Build boards — collect works into persistent local boards (mood boards / reading lists) and export them to a shareable, editorial HTML gallery.

It talks to the live public API at https://everymuseum.org — no database, API key, or local copy of the app required.

Add it to Claude

Claude Desktop

Edit your claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

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

Restart Claude Desktop. You'll see the EveryMuseum tools available.

Claude Code

claude mcp add everymuseum -- npx -y everymuseum-mcp

Running from a local checkout (before npm publish)

Build once, then point your client at the compiled entrypoint:

cd mcp
npm install
npm run build

{
  "mcpServers": {
    "everymuseum": {
      "command": "node",
      "args": ["/absolute/path/to/CulturalAtlas/mcp/dist/index.js"]
    }
  }
}

Claude Code equivalent:

claude mcp add everymuseum -- node /absolute/path/to/CulturalAtlas/mcp/dist/index.js

Try it

"Show me some Hokusai prints." "Tell me more about the third one." "Find works similar to it." "Start a board called Edo seascapes and add those two." "Export the board and open it."

Tools

| Tool | What it does | | --- | --- | | search_artworks | Free-text search; returns captioned results with images. | | get_artwork | Full metadata + high-res image for one work (by id). | | find_similar | Related works for a given artwork id. | | suggest_searches | Curated ideas drawn from the archive's top facets. | | create_board | Create a saved board. | | add_to_board | Add works (by id) to a board. | | view_board | Show a board's works with images. | | list_boards | List all saved boards. | | remove_from_board | Remove works from a board. | | export_board | Render a board to a shareable HTML gallery (optionally open it). | | delete_board | Delete a board (requires confirm: true). |

Boards

Boards are stored locally as JSON at ~/.everymuseum/boards.json and persist across conversations. Each added work is snapshotted, so a board still renders even if the archive changes. Exported galleries are written to ~/.everymuseum/boards/<slug>-<id>.html — self-contained, offline-friendly, and safe to share.

Configuration

| Env var | Default | Purpose | | --- | --- | --- | | EVERYMUSEUM_API_BASE | https://everymuseum.org | Archive to query (set to http://localhost:3000 for local dev). | | EVERYMUSEUM_DATA_DIR | ~/.everymuseum | Where boards and exports are stored. |

Develop

npm install
npm run build      # tsc → dist/
npm test           # build + end-to-end stdio smoke test against the live API

The smoke test (test/smoke.mjs) spawns the server, speaks JSON-RPC over stdio, and exercises every tool against the live archive using a throwaway data dir.

Attribution & licensing

Images and rights belong to the source institutions. The archive only surfaces openly-licensed / public-domain records, but always check the per-work license field before reuse. This server's own code is MIT-licensed.