Hi everyone,
For a while now I’ve been trying to get AI coding agents to build Perspective screens for me. They’re fast, but the view.json they write usually looks fine and then doesn’t work: component types that don’t exist, props in the wrong place, bindings to tags that aren’t there. And the agent never sees the screen, so it has no idea.
So I built a small toolkit called PDK. The agent checks its view against the Perspective schemas from your own install, deploys it to a sandbox project on a dev gateway, and then opens it in a headless browser to see what actually happens: bad tag quality, script errors, text getting cut off, that kind of thing. Then it fixes the problems and tries again.
Here’s what that loop looks like:
It runs as a CLI or as an MCP server, so it should work with Claude Code or any other agent that speaks MCP.
A few things to know up front: it’s meant for development gateways only, it needs Ignition 8.3, and the gateway module is unsigned. It only writes to the one sandbox project you choose.
It’s open source (Apache 2.0) and free to use:
- Repo: GitHub - fcastel2002/pdk: PDK for Ignition Perspective: lets an AI agent build, validate, deploy and inspect Perspective views on a development gateway. · GitHub
- Docs: PDK for Ignition Perspective
I’d love for some of you to try it and tell me what you think, especially what breaks or what’s missing for the way you work. Issues, ideas and PRs are all welcome.
Thanks!
(Just to be clear, this is a personal project and isn’t affiliated with Inductive Automation.)
I know LLMs hallucinate a lot with Ignition; this is my attempt at making the agent check its work against the real schemas and a real gateway instead of guessing.