Paste one line into your AI. Next time it needs something hard to draw — a graph, a chessboard, a music score, a deep-config chart — it fetches a ready-made open-source component and renders it, instead of hand-coding the whole thing.
No install, no MCP, no plugin. The line you paste teaches your AI an endpoint and when to reach for it — it writes that into its own memory.
One line goes into your AI's memory: the endpoint, and the rule for when a visual is worth fetching vs. hand-writing.
When you ask for something hard to draw, your AI calls /api/render?q=… and gets the best-matching open-source component: a data schema + mount code.
Your AI generates just the data and mounts it. Zero component code = the source of reliability. Simple charts, mermaid, forms — it keeps hand-writing those.
Hand-picked from the catalog — each one below is really running from its example data, not a screenshot. This is what your AI mounts.
Type a need below — the same /api/render call runs, and a real open-source component renders live. This is exactly the experience your AI gives you.
The one-line prompt is the easy path. If your client can be configured, or you want zero external dependency, here are the other two.
For config-capable clients — Claude Code, Codex CLI, desktop apps.
claude mcp add --transport http open-visualize https://visualize.openmcp.app/mcp
Gives your AI the tools search_visuals, get_visual, list_scenes.
No API, no dependency. The catalog is open source (MIT).
Browse it straight off the CDN: read discovery/index.md (a light router), pick an id, read data/catalog/<id>.json for its schema + boot. Whole-catalog dump: discovery/llms.txt.