MCP Server & AI

Share your knowledge Ask the community
1

Hi Folks,

I am on the latest Beta of Sketch and started to play around with the MCP Integration. At that moment I really just want to get some simple AI integration to work, to see what is possible.

I got Sketch and Cursor to “work” together but ultimately I would like to use a local model with Sketch. The way I see it LM Studio seems the best option. But i failed to make it work. I can make LM Studio access Sketch via MCP but the model failed to interact with Sketch. I assume that I would need to use a Model that is bigger than what my MacBook Pro can’t handle.

So my question is if someone of you experimented with AI integration and Sketch at all? Did someone try to use a local Modell? What are your general experiences, takeaways?

/Andreas

2

Hello Andreas, what model are you using in LM Studio? Make sure it supports tool calling in order to properly access MCP functions (sketch and others)

3

I am using qwen/qwen3-vl-8b. But i suspect that the model is too small to handle Sketch. When i enable the Sketch MCP my context tokens are immediately larger than the max.

4

it’s the right model IMO, but try to go for the 16b at least, for indeed, you want a smarter model (you can customize LM Studio context window for that model btw, by default it should be 4098, try to go for 32k?)

EDIT, you’re using the VL model, VL model is not great for tools, try the default qwen3 models, not VL.

1 Like
5

I see, my main issue was the context window size. i changed that and the model is working. Thx.

1 Like