MLX MCP Server — Team Setup Guide

1

Prerequisites

setup

Confirm your environment has all required dependencies before installing the server.

Verify Python 3.11 or higher

python3 --version

✓ Validation: Output shows Python 3.11.x or higher.

Verify pip is available

pip --version

Install GitHub CLI and authenticate

brew install gh && gh auth login

Install MLX LM (Apple Silicon) — or install Ollama for Intel/Linux

pip install mlx-lm

⚠ Note: MLX only runs on Apple Silicon (M1/M2/M3/M4). If you are on Intel Mac or Linux, install Ollama instead: brew install ollama

2

Install mlx-mcp-server

install

Install the package from PyPI and confirm the CLI is accessible.

Install the package

pip install mlx-mcp-server

Verify the CLI is working

mlx-mcp-server --help

✓ Validation: Prints usage information including the install subcommand.

3

Configure Claude Desktop

config

Register the MCP server with Claude Desktop so it appears as a toolset.

Preview the config that will be written (dry run)

mlx-mcp-server install --dry-run

✓ Validation: Prints JSON config — confirm it looks correct before writing.

Write the config to Claude Desktop

mlx-mcp-server install

✓ Validation: Prints Added mlx-mcp-server to Claude Desktop config.

⚠ Note: If you are using Ollama instead of MLX, pass additional flags: mlx-mcp-server install --base-url http://localhost:11434 --model mistral

Fully restart Claude Desktop (Quit from menu, then reopen — do not just close the window)

4

Start your local LLM

runtime

Download a model and start the inference server that Claude will talk to.

Apple Silicon (MLX)

Download the Mistral 7B 4-bit model from HuggingFace

pip install huggingface_hub && \
huggingface-cli download mlx-community/Mistral-7B-Instruct-v0.3-4bit

Start the MLX LM server

mlx_lm.server --model mlx-community/Mistral-7B-Instruct-v0.3-4bit

✓ Validation: Logs show Starting server on http://localhost:8080.

Ollama (Intel Mac / Linux)

Start Ollama and pull the Mistral model

ollama serve & && ollama pull mistral

5

Verify in Claude Desktop

test

Confirm Claude Desktop can see and use the MLX MCP tools.

Open Claude Desktop and look for the tools icon in the toolbar

✓ Validation: mlx is listed in the tools panel with 4 tools available.

Run a health check by asking Claude the following

Use the health_check tool to verify my local MLX server is running.

✓ Validation: Claude reports status: ok.

Run a quick inference test

Use the quick_test tool with test_type=math

✓ Validation: Returns a math response including tok/s throughput metrics.

6

Configure Claude Code (optional)

config

Also wire up mlx-mcp-server for the Claude Code CLI, if you use it.

Register with Claude Code

mlx-mcp-server install --claude-code

✓ Validation: Prints Added mlx-mcp-server to Claude Code config.

Verify the tools are visible from Claude Code

Use the list_models tool

✓ Validation: Claude Code lists available MLX models.

7

Share with your team

team

Spread the word — two commands to get any teammate running.

Share the one-liner install commands

# Apple Silicon (MLX)
pip install mlx-mcp-server && mlx-mcp-server install

# Intel Mac / Linux (Ollama)
pip install mlx-mcp-server && mlx-mcp-server install --base-url http://localhost:11434 --model mistral

Point teammates to the repo for docs and issue tracking

https://github.com/deresolution20/mlx-mcp-server