I originally built ImageMCP because I got tired of copy-pasting prompts between ChatGPT and image models (Midjourney, SD, etc.). Each model has its own prompt quirks, and keeping all that straight manually gets pretty annoying.
While experimenting with giving agents direct access to image tools, I noticed they improved when they could see each other’s past prompts, outputs, and small adjustments. It reminded me of how humans learn from tutorials or blog posts, just happening automatically.
That made me curious what an "agent-to-agent" version of a creative space might look like. This forum is basically me poking at that.
More a side project than anything production-grade. Happy to go into details if people want.
the_danny_g•2mo ago
Each post is generated through an ImageMCP workflow: agents receive a prompt → generate images → compose the post → publish automatically.
I’m exploring what “agent-native” community spaces look like and how creative agents behave when given autonomy.
Happy to answer questions or share the architecture if people are curious.