local_vision

A native visual analysis tool based on MCP (Model Context Protocol) that uses LM Studio's vision model to analyze images.

• 🔍 Local Image Analysis: Supports analyzing images in local PNG, JPG, JPEG, WEBP formats
• 🤖 Visual Model Integration: Integrate LM Studio's vision models for image understanding
• 🛡️ Security Restrictions: Only allow access to locally existing image files to prevent path traversal attacks
• ⚡ Fast Response: Asynchronous processing with support for custom analysis prompts

• Python 3.8+
• The LM Studio(or other local server) service is running
• Supported image formats: PNG, JPG, JPEG, WEBP

Make sure Conda is installed and create an 'mlx' environment:

conda create -n mlx python=3.10
conda activate mlx

To install the required dependencies in the project directory:

pip install mcp httpx

1. Launch LM Studio
2. Download and load a vision model (e.g. 'qwen3.5:2b-bf16')
3. Make sure the LM Studio API service is running at 'http://localhost:11434'

The following environment variables can be set up:

# Set the LM Studio service address (default: http://localhost:11434)
export LM_STUDIO_URL="http://localhost:11434"

# Set the visual model name (default: qwen3.5:2b-bf16)
export VISION_MODEL="qwen3.5:2b-bf16"

# Set the Conda environment name (default: mlx)
export MCP_CONDA_ENV="mlx"

1. Direct Run:

python main.py

1. Configure in Opencode:

Add in your ``opencode.json``:

{
 "mcp": {
    "local_vision": {
      "type": "local",
      "command": ["python", "your path"],
      "environment": {
        "LM_STUDIO_URL": "http://localhost:11434",
        "VISION_MODEL": "qwen3