41 recipes for turning AI agents into media studios

What it does

This repo is a collection of 41 workflow recipes and thin CLI wrappers that plug AI agents into commercial generative-media APIs. Agents read SKILL.md files, follow declared input schemas, and execute multi-step pipelines—turning a product photo into a cinematic 10-second ad, or a long video into ranked vertical Shorts—by calling out to muapi-cli.

The interesting bit

The architecture is deliberately split: /core holds minimal API plumbing, while /library bakes in domain expertise—“Cinema Director” for shot vocabulary, “Nano-Banana” for reasoning-driven image generation, “AI Clipping” for server-side virality ranking without local Whisper. It’s prompt engineering as structured infrastructure, not just better adjectives.

Key highlights

• 41 end-to-end recipes across video (16), social (5), and visual/design (21) categories
• MCP server exposes 19 tools directly to Claude Desktop, Cursor, or any MCP-compatible agent
• Claims access to 100+ models including Midjourney v7, Flux Kontext, Seedance 2.0, Kling 3.0, Veo3
• --view flag auto-downloads and opens generated media; --jq filtering for agent-friendly JSON extraction
• Local file auto-upload to CDN for processing

Caveats

• All generation flows through muapi-cli, a proprietary npm/pip package tied to muapi.ai; this is not self-hosted
• The repo itself is mostly glue code, documentation, and prompt templates—substantial work happens behind the API
• Several skills note endpoints “don’t yet have a CLI alias,” implying some recipes require raw API fallback

Verdict

Worth a look if you’re already living in Claude Code or Cursor and want pre-packaged creative workflows without hand-rolling prompts. Skip it if you need offline operation, transparent pricing, or fine-grained model control—the muapi.ai dependency is the whole stack.