The archive, for agents.

Arkivism runs a hosted MCP server — connect your AI assistant and it can search a million-plus public-domain images the same way you do: by describing them. Every result carries its licence and a ready-to-paste credit line. No API key, free.

Connect

The endpoint speaks MCP over streamable HTTP — one URL, no key, works in any MCP-capable app:

https://api.arkivism.com/mcp

Claude Code

claude mcp add --transport http arkivism https://api.arkivism.com/mcp

Claude Desktop / claude.ai — Settings → Connectors → Add custom connector, paste the endpoint URL.

ChatGPT — Settings → Connectors → Create (developer mode), server URL as above, no authentication.

Cursor — add to .cursor/mcp.json (or Cursor Settings → MCP):

{ "mcpServers": { "arkivism": { "url": "https://api.arkivism.com/mcp" } } }

OpenAI Codex CLI — in ~/.codex/config.toml:

[mcp_servers.arkivism]
url = "https://api.arkivism.com/mcp"

Gemini CLI — in ~/.gemini/settings.json:

{ "mcpServers": { "arkivism": { "httpUrl": "https://api.arkivism.com/mcp" } } }

VS Code (Copilot agent mode) — in mcp.json:

{ "servers": { "arkivism": { "type": "http", "url": "https://api.arkivism.com/mcp" } } }

Anything else (Windsurf, Zed, LibreChat, your own agent…) — any MCP client that speaks streamable HTTP works; point it at the endpoint. If your client's config format differs from the samples above, check its MCP documentation — the URL is all it needs.

What it can do

• search_archive — natural-language semantic search ("dramatic stormy sea with ships"), with date, medium and result-count options.
• find_similar — visually similar works to any item; explore outward from a good result.
• get_item — one item's full record: metadata, licence, rights URL, attribution, full-resolution link.
• archival_echo — reverse-image search: give it an image URL, get the closest visual matches in the archive.
• get_download_url — the full-resolution file for an item, with the credit line to carry along.

House rules

Everything indexed is CC0 or public domain — safe for any use, including commercial. Include each item's attribution text when you use an image: not legally required, but it's the norm we uphold toward the museums and archives that opened their collections. Polite rate limits apply; if you're building something real on the archive, write us.

Prefer plain HTTP?

The underlying JSON API is documented in llms.txt — stable paths, same data.