4 hours ago
Hey everyone 👋
Recently I've been building a bunch of MCP servers with Workato, and somewhere along the way I got hooked on using AI coding agents for everything. Claude Code, Cursor — once you start, it's hard to go back to doing things manually.
That got me thinking: why am I still clicking through the Workato UI to build recipes? So I started trying to use coding agents for recipe development too — and immediately ran into a wall. They'd hallucinate action names, get the JSON structure wrong, or confidently suggest connectors that don't exist.
So I built a toolkit to fix that: workato-dev-kit — a context layer that gives AI coding agents actual knowledge of Workato's recipe format, connector behaviors, and platform features.
https://github.com/rkawaishi/workato-dev-kit
It works as a git submodule on top of your workspace repo. Once set up, you can use slash commands like `/create-recipe`, `/create-connector`, or `/create-genie` and the AI actually produces valid Workato JSON that you can push with the Platform CLI.
It covers:
- 316 connectors with documented actions/triggers
- Recipe JSON generation
- Workflow App and MCP server scaffolding
- A learning cycle that improves from your own pulled recipes
I've been using this in real projects and honestly it's been working pretty well — recipes come out cleaner than I expected.
That said... AIRO is coming. Once that's out of private preview, this whole thing might be obsolete 😅 But until then, happy to share what I've got.
Curious if anyone else here has gone down this rabbit hole — using Claude Code or Cursor alongside the Platform CLI. Would love to compare notes.