cull

What it does

Wraps the Cull image-library app into a usage map: import folders, browse, rate and decide, build manual and smart collections, run vision and quality analysis, find visually similar images via embeddings, and export or publish. The default path is the cull CLI plus the cull:// URL scheme — no MCP connection required and restart-proof — with the mcp__cull__* tools reserved for richer interactive control.

Key features

• Review in Cull, never Preview — the core rule: display by importing the folder and fronting the app (cull import_folder + open "cull://open?path=…&view=grid"); never open image files.
• CLI-first, MCP optional — the headless binary writes to the same library DB the running app reads; the MCP is the fallback for live navigation, rating round-trips, and find-similar (and it drops on app restart).
• Curate fast — rate, set keep/reject decisions, and gather favorites into manual or smart collections.
• Find-similar via embeddings — surface visually related images (download the embedding model and generate embeddings first).
• Vision & quality passes — object detection, search-by-object, and quality scoring across a whole library.
• Export & publish — export presets, static publish packages, and clipboard publishing.

How it works

The cull binary runs headless subcommands (import, list, export, embeddings, quality) that share one Rust core with the GUI, URL scheme, and MCP server. MCP tools are named mcp__cull__* and are deferred in Claude Code, so they are loaded with ToolSearch before calling; long operations (embeddings, analysis) are async jobs to poll.

When to use

When you want to view, review, rate, organize, search, or export images / AI-art generations with the Cull app — defaulting to the CLI and URL scheme, with the MCP available for interactive curation.