opencode-minimax-easy-vision

Automatically intercepts images pasted into the OpenCode editor via clipboard events, saves them to a persistent local filesystem directory with timestamped filenames, and maintains a registry of saved image paths. The plugin hooks into the editor's paste event lifecycle to detect image data (via DataTransfer API), serializes it to disk, and tracks references for downstream MCP tool injection.

Dynamically generates and injects MCP (Model Context Protocol) tool definitions into the Minimax model's system prompt, exposing saved image file paths as callable tool parameters. The plugin constructs a JSON schema describing available images and their metadata, then wraps this schema in MCP tool format (following the OpenAI/Anthropic function-calling convention) so the Minimax model can reference images by path when generating responses.

Provides native bindings to Minimax's vision-capable model endpoints, handling authentication via API keys, request formatting for vision inputs (image paths + text prompts), and response parsing. The plugin abstracts the HTTP/REST layer, managing session state and model selection so developers can invoke vision analysis without directly constructing Minimax API calls.

Implements the OpenCode plugin architecture lifecycle, including initialization hooks, event registration (paste events, model selection changes), configuration loading, and cleanup on plugin unload. The plugin registers itself with OpenCode's plugin manager, declares its capabilities via a manifest, and responds to editor lifecycle events to activate/deactivate vision features.

Maintains an in-memory registry of all pasted images, storing metadata (filename, save path, timestamp, dimensions, file size) and providing query/lookup methods for downstream components. The registry is keyed by image path and supports filtering by timestamp or metadata attributes, enabling the MCP tool injector to enumerate available images and the UI to display image history.