[Feature]: Native support Node-RED HA addon

[vsisl]

Problem / Use Case

Hello,

Thank you for this great project!

I would appreciate native support of the MCP to interact with Node-RED home assistant addon. Node-RED is great for advanced automations so I believe others would appreciate this feature as well.

Proposed Solution

Extend the MCP to be able to directly work with Node-RED.

Alternatives Considered

No response

Additional Context

Already now, Claude Code (and likely other agents too) us able to interact with Node-RED through curl GET and POST commands. However, native support would be beneficial.

[Coolver]

@vsisl Oh, that’s a really cool idea – thanks for suggesting it
I also think a lot of people would love tighter Node-RED + MCP integration for advanced automations.

From my side I see two main ways to approach this:

1. Go via Home Assistant (indirect control) Use whatever Node-RED exposes through Home Assistant: call HA services that trigger Node-RED flows, maybe work with entities that Node-RED creates/uses, stay inside the existing HA security model.

This would be simpler to set up (no extra URLs/tokens), but the feature set would be limited to what Node-RED exposes via HA, not its full capabilities.

2. Talk to Node-RED directly (admin API)
   Use the Node-RED admin API from the agent: list / read / update flows, deploy changes, possibly inspect nodes, subflows, etc.

This would be much more powerful, but: requires exposing the Node-RED admin API to the agent, and adding some config to the HA Vibecode Agent UI (Node-RED URL, auth token, etc.), plus some extra work on my side to make it safe and reasonably easy to configure

From your perspective as a Node-RED user which approach would you actually prefer? Are you thinking about: listing flows, editing them from Cursor / VS Code, deploying, or something more specific (e.g. “given my HA entities, generate a new Node-RED flow that does X”)?

My guess is that option 2 (direct Node-RED API) is what you really want, but I’d love to confirm that and understand your ideal workflow before I start designing the tools

[vsisl]

Hi @Coolver

I think option 2 is the better solution.

What I imagine as an ideal solution is that the ai agent would be able to interact with node red on the level of individual nodes. E.g. explain what a node does, add a new node etc.

Right now, as far as I understand, the agent is only able to interact on the level of the whole flow. So it can create a new flow or replace an existing one by overriding it completely. It does that through the curl POST commands I've mentioned.

I hope this helps. Thanks again for your great work!