Turn Your Coding Agent Into a Multi-Model Design Studio

What it does

MeiGen exposes eight tools through the Model Context Protocol, turning any MCP-compatible coding assistant into a design operator. It can search a curated library of 1,446 prompts, enhance bare ideas into detailed descriptions, and route generation requests across cloud APIs or a local ComfyUI instance. Images and videos are saved locally; reference files are auto-compressed and uploaded before generation.

The interesting bit

Instead of blocking the main conversation while a model renders, MeiGen spawns isolated sub-agents—image-generator, prompt-crafter, and gallery-researcher—to draft prompts and run batches in parallel. The main agent stays clean, and the sub-agents even run on cheaper Haiku models to keep costs down.

Key highlights

• Nine models including GPT Image 2, Midjourney V8.1, and Flux 2 Klein, plus a local ComfyUI backend, with automatic routing
• 1,446 curated prompt templates drawn from the nanobanana-trending-prompts library
• Runs fully offline when pointed at a local ComfyUI node; cloud generation requires a MeiGen or OpenAI-compatible API key
• Standalone CLI mode for shell scripts and CI pipelines, independent of any MCP host
• macOS auto-open hook that launches generated images in Preview

Caveats

• Video generation and first-time npx downloads can take 5–10 minutes, which exceeds the default timeouts in some MCP clients like Hermes Agent
• Actual image and video generation require an API token; search, prompt enhancement, and model listing are free

Verdict

Worth a look if you want your coding agent to handle visual assets without leaving the terminal. Skip it if you already have a dedicated design workflow that doesn’t need orchestration through an AI chat interface.