Imagine Mcp

A production-grade bridge to image and video understanding plus generation across Gemini, OpenAI, and Grok. You get two core tools: `understand` for reasoning over image and video URLs with configurable context length, and `generate` for text-to-image, image-to-image, and text-to-video workflows. Each provider exposes "poor" and "rich" tiers so you can trade speed for quality on the fly. The server ranks models weekly against Artificial Analysis and LMArena leaderboards, caches responses to disk with configurable TTL, and degrades gracefully when credentials are missing. Ships with stdio and HTTP transports, plus a config tool that surfaces relay forms, credential state checks, and runtime knobs like log level and default provider. Reach for this when you need multimodal ops without hardcoding a single provider.