Expose Firefly generation (image + video) via MCP/API so it can be driven from Claude and other AI agents

Adobe already ships an MCP connector that lets Claude reach Creative Cloud — asset storage, Lightroom-style adjustments, background removal, vectorise, InDesign merges, video render. It's genuinely good. But the one thing it can't do is the thing Firefly is actually for: generate an image or a video from a prompt.

That means my workflow is split down the middle. I plan a sequence with Claude — shot list, prompt text, aspect ratios, continuity notes — and then I have to hand-carry every prompt into the Firefly web UI one at a time, download the outputs, re-upload them, and manually keep track of which frame is which. For a 7-episode vertical micro-serial, that's hundreds of context switches for work the agent could already orchestrate end to end.

Higgsfield shipped an MCP server this spring that does exactly this, and it's now the default entry point for a lot of us: connect once, describe the shot, the agent picks the model, fires the generation, polls, and returns the clip in chat. It works from Claude web, desktop and Claude Code. No API key — OAuth through the account you already pay for. The result is that Firefly, which has better provenance, better integration with the rest of the Adobe stack, and a licensing story no one else can match, is losing agent-driven work to a competitor purely on plumbing.

What I'm asking for:

1. Text-to-image and text-to-video generation tools in the existing Adobe MCP server — same auth, same asset storage, outputs landing straight in Creative Cloud Files.
2. Availability on individual Creative Cloud / Firefly plans, not enterprise-only Firefly Services contracts. The people building this way are mostly solo creators and small studios.
3. A CLI for terminal-based agents, so the same generations can run inside scripted pipelines.
4. Batch and reference support — style/character reference images, multi-shot batches, seed control. Consistency across shots is the whole game for narrative work.
5. Content Credentials preserved on every asset returned through the API. This is Adobe's genuine differentiator, and it should be the default, not something I have to reattach later.

Why this is worth building: agent-orchestrated generation isn't a niche. It's becoming the way serialised and high-volume creative work gets made. Adobe already has the models, the storage, the editing tools, and the connector. The generation endpoints are the missing link, and without them the connector reads as a nice editing utility rather than a production environment.