Cline
Cline CLI and the VS Code extension can use Rapid-MLX through an OpenAI-compatible /v1/chat/completions connection.
Quick start
$ npm install -g cline $ rapid-mlx serve qwen3.5-4b-4bit $ rapid-mlx agents cline --setup $ cline --yolo "Read a file and explain it"
Run the server in one terminal, then setup and Cline in another. Setup detects the served model, previews its changes, asks before writing, and checks the connection. Use --yes in a script.
Where setup writes
The CLI reads ~/.cline/data/settings/providers.json. The openai-compatible provider entry holds baseUrl, apiKey, and model. Setup preserves other providers and backs up an existing file. For a server on another port, add --base-url http://127.0.0.1:18730/v1. For a protected server, export RAPID_MLX_API_KEY first.
CLINE_DATA_DIR changes Cline's data root; CLINE_DIR changes its parent; CLINE_PROVIDER_SETTINGS_PATH overrides the provider file. rapid-mlx launch cline --server-url http://127.0.0.1:18730 is an alternative setup path.
VS Code
If a Cline session already selected another provider, open Cline → Settings → API Configuration. Select OpenAI Compatible, set the Rapid-MLX /v1 URL and served model ID, and use the same API key. The CLI provider file does not force an open editor session to switch models.
Check the connection
rapid-mlx agents cline --test runs wire and CLI checks when the binary is installed. For a tool check, ask Cline to read a local file and verify that its reply contains the file's actual text.