I want to share a short retrospective on trying to use Penpot as part of a real AI-assisted workflow with Codex, MCP, local backend access, reusable components, and model-to-projection work.
This is my own research and opinion from one serious integration attempt, not an attempt to speak for Penpot or for every user workflow.
Why I was trying it:
- I wanted Penpot to act as a live visual projection surface for Event Modeling, screen paths, and reusable UI/state work
- I was interested in the bridge between design, modeling, AI manipulation, and later generated application surfaces
- I was hoping for not only
model -> projection -> artifact, but alsopenpotDesign -> ingestion -> model, where an existing design could feed back into modeling and real app production - I was specifically trying to see whether Penpot could function as more than a passive design artifact in this workflow
What worked:
- backend access and export were useful
- Penpot was valuable as design evidence and a projection surface
- some live board/component work did work
Why I paused it for now:
- the workflow stopped being deterministic enough for the phase I’m in
- MCP transport/connectivity was fragile in practice
- plugin/API behavior diverged
- reusable instance overrides and some render/export paths were not reliable enough for the active workflow
So this is not a “Penpot is bad” post.
It is more: Penpot was useful research, but it became more integration friction than modeling leverage for this phase, so I moved the active work to deterministic F#/HTML/CSS renderers and kept Penpot as supporting evidence rather than the primary engine.
I wrote up the deeper details here:
- Deep retrospective: https://github.com/IvanTheGeek/NEXUS-EMERGING/blob/main/docs/research/penpot/codex-mcp-retrospective.md
And I already filed or linked narrower aspect posts separately, so I’m not trying to rehash all of that here:
- Existing GitHub issue: https://github.com/penpot/penpot/issues/8826
- Existing forum plugin note: https://community.penpot.app/t/what-s-working-and-what-s-missing-in-penpot-plugins/9785/8?u=ivanthegeek
- Existing forum text-rendering note: https://community.penpot.app/t/penpot-bug-report-on-text-rendering-overrides/10460?u=ivanthegeek
I also think this older thread is worth keeping in view as trace:
If the Penpot team or other users are interested, the deeper note includes:
- exact local Codex log references
- the transport failure trail
- the motivations for trying Penpot this way
- the deeper workflow problems that made me pause it for now
