Files list too large for context window

[remcoder]

Hi there! Thanks for making this awesome project available!

I've just started playing with it and I'm running into the issue that when I request a list of files, the resulting data is too large for the context window.

I was using:

• the Octoprint tool
• VS Code + Cline
• DeepSeek v3 (which has a 64k input window)
• the json result from Octoprint was 500kb (324 files, ~200k tokens)

I would think that a subset of the data per file would suffice in many cases, something like: name, data, size. This would be less than 10 times of the data, so less likely to hit the limit. So that is what I would suggest.

[DMontgomery40]

Are you open to using Gemini Max? The new Gemini 2.5 Max just got updated to handle tool calling much better including MCP, and has a theoretical context window of 1M tokens. In reality, it's more like half that, but still pretty damn good. Also, it's the best coding LLM out, as of the writing, across the board.

[remcoder]

well for regular use I don't want to keep throwing money at it 😄 That's what DeepSeek solves for me atm. The next step would be to have a locally running model (on my laptop) to solve not just cost but also privacy and IP concerns. And hopefully at some point in the future we can run a decent tool-capable model on a Pi.

In my view, that is the overall direction for MCP tools that are coupled with open source software like OctoPrint. For day-to-day programming, sure, we can pay for a more powerful model because it's a positive ROI.

Of course for now, to get on with playing with this project, I can use Gemini 2.5 Max, as you say. Thanks for the suggestion!

Personally, my request would be to keep this ticket open to for the time being with the intention of reducing the output from the tool as an optimization.

How do you feel about this? Or would such an optimization be out of scope for now?