Designing animation asset galleries for agents

Visual asset galleries are often designed only for humans. Cards look attractive, but names change, downloads are hidden behind JavaScript, and an agent cannot tell whether a file is reusable or what license applies.

The app.nz animation commons now treats the gallery and the agent interface as two views of one index. Motion, VFX, terrain, procedural objects, pose backends, and image-to-3D examples all expose the same stable IDs and metadata.

One search and download contract

GET /api/animation-library/index.json returns the cacheable full manifest. Use q, kind, and limit for focused search. Every asset supports a JSON recipe; terrain and 3D objects also support OBJ geometry.

MCP clients use four narrow tools:

• search_animation_library finds assets and preserves source/license fields.
• download_animation_asset returns a URL or inline JSON/OBJ content.
• create_image_to_3d_recipe prepares a reconstruction job.
• generate_terrain_recipe produces deterministic worlds.

The same operations exist in the Go CLI and desktop MCP host. This matters for coding agents: the workflow remains searchable and testable whether it runs in a browser, terminal, desktop app, or remote tool server.

The implementation is split into four MIT repositories: motion-kit, vfx-kit, terrain-kit, and object-kit.