Claude integration for Moonraker API.

[obsidian fractal]

I create an integration (MCP) for claude.ai that integrates it into Moonraker. the source is available on github. https://github.com/TyanColte/moonrakermcp

Prompts:

Give me a detailed printer status in a nice interface.

Result: <Image 1>

once the bed heats up, I want you to run an probe_eddy_ng_tap and report the tap result using your new log reader. make the interface something similar to what you did earlier. Then do a eddyng experimental bed level. if you can pull the points of the scan, generate a graph of the bed. then set the toolhead back to center with the height at 25mm

After some time: Claude reported back that the axis were homed and did a tap test, result: <image 2> then as part of the same prompt, scanned the bed, and created a bed mesh with detailed information <image 3> set the toolhead back to center at z25, and is waiting for further instructions.

It also recommends actions based on the results it recevies and displays.

Quick read on the bed: range is about 0.21mm peak-to-peak, with a clear front-to-back tilt — the front of the bed is high (especially front-right at +0.123) and the back-right corner is the low spot (-0.083). Mesh stddev is 0.046mm, which is workable for printing but suggests the bed could use a tramming pass if you want to flatten it further. Want me to cool things down, run PROBE_BED_TRAMMING to get screw adjustment recommendations, or something else?```

[red cove]

"oh yeah lets give ai unrestricted access to a machine that can start a fire" you better have some crazy failsafes on this

also whats the usecase for this? i dont see anything this can do that cant be done more safely and consistently with a macro

[obsidian fractal]

Being able to communicate in a more natural way to your printer, and having a LLM with access to the vastness of the internet assist with analyzing, calibrating, and troubleshooting.

[red cove]

your first point is literally just what mainsail is for, and your second point is just throwing out buzzwords without explaining anything. also you didnt address the saftey concerns i had.
I think i can speak for everyone when i say that we dont want the thing thats destroying foss in fosh

[red cove]

^

[obsidian fractal]

I think you're overestimating the amount of autonomy these LLMs have, it can't do anything you don't tell it to do, and it won't do anything without prompting. It's not a constant connection to the printer. You have to request it to take any action and Claude will confirm any action that it performs before it starts. And it will not override any of the built-in safeties that 99% of printers already have. (thermal runaway, un-homed axis, etc.) Being able to speak in full sentences to an LLM and have it perform tasks on the printer, batching multiple tasks, and quickly analyzing results of tests that output large amounts of obscure data is the goal with this.