Run Local Coding Agents on Mac with MLX Models

A local MLX server and agent-specific setup commands for coding on your Mac.

Run rapid-mlx serve qwen3.5-9b-4bit on your Mac. In another terminal, choose a command from the coding agent setup guides, such as rapid-mlx agents codex --setup. The served model runs behind local OpenAI and Anthropic API routes.

brew install rapid-mlx codex
rapid-mlx serve qwen3.5-9b-4bit  # keep this terminal running
# In another terminal:
rapid-mlx agents codex --setup
codex

The 9B model is the Codex guide's practical floor for a 16 GB Mac. A larger model needs more unified memory. Use rapid-mlx recipe to see a model suggestion for your machine, and check the hardware tiers before serving larger weights.

Pick the agent you use

Each guide shows the setup command, the client command, the file or environment variables changed, and how to test the connection. Cline also has a VS Code extension path. Aider's setup prints exports that you paste into the shell before running Aider.

Use the right base URL

OpenAI-compatible clients use http://localhost:8000/v1. Claude Code uses http://localhost:8000, because its client adds /v1/messages. The server prints its actual port on startup; if port 8000 is busy, use the printed address. The local MLX server guide has a curl request and the available routes.

The integration matrix records agent-specific verification and known failures. Model capability still matters for code edits and tool use, so start with a model suited to your Mac's memory.

curl -fsSL https://rapidmlx.com/install.sh | bash