Hi Brian,
I don't know specifically on a Mac nor in what framework you're planning to use the server, nor do I have Claude that seems to be expected by the installation instructions in the J-MCP repo.
So, probably, the following is not going to help your case so much, but perhaps it's interesting for someone else...
I did get the J-MCP server to work with KIT (
https://github.com/mark3labs/kit) using local Ollama models (Gemma4:26b) seems to work rather well, even locally on a laptop with only an iGPU (Using Vulkan).
For getting J-MCP to work with KIT, you'd have to copy the binary into the bin directory in the J installation, and configure a local MCP server by putting this in the configuration file (.kit.yml by default):
mcpServers:
j-mcp:
type: "local"
command: "<path to J install>/bin/j-mcp-linux-x86_64"
args: []
enviroment:
JPATH: "<path to J install>/bin/jconsole"
For those wanting to use Vulkan with Ollama on Linux, (e.g. unsupported GPU for NVidia Cuda or AMD ROCm), I found it works great even with a simple integrated GPU, giving a considerable speed-up compared with CPU-only. It requires some configuration of Ollama using sudo systemctl edit ollama, adding:
[Service]
Environment=HIP_VISIBLE_DEVICES=-1
Environment=OLLAMA_VULKAN=1
Environment=GGML_VK_VISIBLE_DEVICES=0
Environment=OLLAMA_IGPU_ENABLE=1
Good luck,
Jan-Pieter