n8n-nodes-typesafe-jev: Typed Decisions Inside No-Code Workflows
A small n8n community node lets no-code workflows call Jev for classification, routing and scoring. Here's what it ships, what it needs, and where it's still rough.
Everything covered elsewhere in this series wires Jev into a coding agent or an editor. n8n-nodes-typesafe-jev goes somewhere different: it's a community node for n8n, the open-source workflow automation tool, that lets a no-code pipeline call Jev directly. That's a smaller and less flashy audience than "your Claude Code session," but it's arguably a more natural fit — n8n workflows already branch on conditions, and Jev exists specifically to answer typed conditional questions fast and cheap.
At 2 stars as of September 2026, this is one of the newest and least-established projects in the batch. It's worth covering anyway, because it's a clean, minimal example of what "Jev as a workflow primitive" looks like outside the agent-development audience this series otherwise targets. For what Jev itself does, see what Jev is and typed decisions vs free text.
Key Takeaways
- It's a thin, purpose-built wrapper, not a general MCP bridge — it exposes Jev specifically as an n8n node, with n8n's own form-building UI standing in for a schema editor.
- Two configuration paths. Build questions one at a time in the node's form UI, or supply a complete JSON schema from an upstream node for reusable decision logic.
- It declares itself usable by n8n's AI Agent nodes, not just as a static workflow step — an agent node can supply state and receive a typed decision instead of freeform text.
- Setup is minimal: a TypeSafe API key and Node.js 20.15+, no other runtime dependencies.
- The README flags its own rough edges, including a documented ESLint-fixer bug that corrupts the question-properties file if run carelessly.
What the node does
The node's job is to submit one or more typed questions — Choice, Yes/No (the noul primitive under a friendlier label), or Score against a defined rubric — about the data flowing through an n8n workflow, and return Jev's typed answers back into that same flow. A workflow can, for example, take an incoming support ticket, ask Jev whether it's urgent and which team it belongs to, and route the ticket down different branches based on the typed response, with no LLM call and no prompt to write.
Two configuration modes cover different use cases. The form-based mode is meant for building a decision by hand inside the n8n editor: pick a question type, label its options or rubric levels, give it a unique name. The JSON mode is meant for reuse — an upstream node can construct a full question schema programmatically and hand it to the Jev node as data, which is the more natural pattern once you have more than a handful of decision points or want the same schema shared across multiple workflows.
The node also declares AI Agent compatibility, meaning it can register as a callable tool for n8n's own agent nodes rather than only running as a static step in a linear pipeline. In that mode, an n8n agent supplies state and gets back a typed decision instead of free text, mirroring the same tool-call pattern used by the MCP wrappers — just inside n8n's own agent framework instead of Claude Code or Codex.
What it needs, and what's still rough
Setup requires a TypeSafe API key from the TypeSafe console and Node.js 20.15 or newer; the README states there are no other runtime dependencies beyond n8n itself, which keeps the install surface small. Before any request goes out, the node validates the question schema locally, so malformed configuration is caught before it burns an API call — a small but useful detail for a no-code audience less likely to be reading error logs closely.
The README is upfront about at least one real rough edge: running the project's ESLint auto-fixer corrupts the question-properties file, so formatting there needs to stay manual for now. It also states plainly that this is an unofficial, community project "not affiliated with, endorsed by, or supported by TypeSafe" — worth keeping in mind given how much of this series covers first- and third-party integrations that sound similarly official but aren't.
Given the project's age and star count, treat it as exactly what it is: a working proof that Jev fits cleanly into a no-code workflow tool, built by one person, days old. It hasn't had the kind of community exercise that would surface edge cases beyond the one the author already found.
What to do next
- Try it on a branching decision you already automate manually — ticket triage, content classification, or simple routing — rather than a complex multi-step pipeline first.
- Use JSON mode once you have more than a couple of questions. It's the path that scales, and it keeps the schema in version control alongside the rest of your workflow definition.
- Avoid the ESLint auto-fixer on this repo until the documented bug is resolved, or review its diff carefully if you must run it.
- Compare against a full MCP or ACP bridge like Jevbridge if you need Jev inside both a coding agent and a no-code tool — this node only covers the n8n side.
- Check the radar for other integrations. The Awesome Jev projects list is the place to see what else has shipped since. On aiskill.market: /skills/n8n-nodes-typesafe-jev-n3ndor.