# MCP server for Node-RED

**URL:** <https://discourse.nodered.org/t/mcp-server-for-node-red/101352>\
**Category:** Share Your Projects\
**Tags:** docker, llm, ai\
**Created:** [24 June 2026 17:58 UTC](https://discourse.nodered.org/t/mcp-server-for-node-red/101352 "2026-06-24T17:58:07Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![gmag11](https://sea2.discourse-cdn.com/flex026/user_avatar/discourse.nodered.org/gmag11/32/477_2.png) [@gmag11](https://discourse.nodered.org/u/gmag11)\
**Post date:** [24 June 2026 17:58 UTC](https://discourse.nodered.org/t/mcp-server-for-node-red/101352/1 "2026-06-24T17:58:07Z")

</div>

Hi everyone! I’ve been developing an MCP server designed to fully control and manage any Node-RED instance directly from an AI agent. I just published it on NPM: [https://www.npmjs.com/package/@gmag11/nodered-mcp-server](https://www.npmjs.com/package/@gmag11/nodered-mcp-server)

You can find the setup instructions right in the README.  
⚠ **PLEASE DO NOT USE IT IN PRODUCTION INSTANCES.** ⚠

If you give it a try, I’d love to hear your thoughts and feedback!

You can use it to build flows from scratch using pure natural language, group nodes in an existing flow, or automatically generate documentation/comments for those old _function nodes_ you wrote months ago and no longer remember what they do. 😁

_Again, if you absolutely must test it near anything important, at least back up your data directory first._

**A few quick tips for testing:**

- **Docker:** I highly recommend running it as a container. There is a `docker-compose` example in the repository to spin up Node-RED and the MCP server together. Using network transport, server exposes a viewer to see what LLM is doing in realtime.
- **Quick test:** If you just want to take it for a quick spin, you can use the `npx` method detailed at the end of the README.
- **LLM Compatibility:** I've done _a lot_ of testing. It performs great with large models—it goes quite well with Sonnet 4.6, but also with Deepseek v4 Flash, which is practically free. I haven't tested it with local models. I imagine that with something like Qwen 3 30b or larger, it will do well enough for simple tasks.

It uses Node-RED's Admin API along with some Editor API endpoints. In my testing, it’s been quite stable; it used to occasionally leave a property blank (causing that familiar red warning triangle on the node), but that should be fully fixed in the latest release.

Check it out and let me know how it works for you!