Simple Vision MCP

A lightweight, focused Model Context Protocol (MCP) server designed specifically for image analysis using OpenAI-compatible APIs. Built with TypeScript and the MCP SDK.

Motivation

When working with AI coding agents that don't natively support vision capabilities, you often need a reliable way to analyze images. Many existing MCP vision servers are tightly coupled to specific providers (like OpenRouter or OpenAI) or come with unnecessary complexity.

Simple Vision MCP was created to solve a specific problem: enabling any OpenAI-compatible API endpoint to function as a vision analysis backend. It focuses on doing one thing exceptionally well - analyzing images - while remaining flexible enough to work with any OpenAI-compatible provider.

The Problem We Solved

During setup, we encountered several issues:

1. Many vision MCP servers only support specific providers (OpenRouter, OpenAI, etc.)
2. Container-based solutions had stdio communication issues
3. Python-based servers had dependency conflicts
4. Existing solutions were overly complex for the basic need

Simple Vision MCP addresses these by:

• Supporting any OpenAI-compatible API endpoint
• Running as a native Node.js process (no containers needed)
• Minimal, focused codebase that's easy to debug and maintain
• Zero external dependencies beyond the MCP SDK

Features

• OpenAI-Compatible: Works with any API that follows the OpenAI chat completions format
• Single Tool Focus: One purpose - image analysis done right
• TypeScript: Full type safety and modern JavaScript
• Minimal Dependencies: Only essential dependencies
• STDIO Communication: Native MCP protocol support
• Configurable: Full control via environment variables
• npx Support: Can run directly with npx, no installation required

Installation

Prerequisites

• Node.js 18 or higher
• An OpenAI-compatible API endpoint with vision capabilities

Quick Start with npx (Recommended)

No installation required - just run directly:

npx -y @erickstryck/simple-vision-mcp

Global Installation

npm install -g @erickstryck/simple-vision-mcp

From Source

git clone https://github.com/erickstryck/simple-vision-mcp.git
cd simple-vision-mcp
npm install
npm run build

Configuration

Simple Vision MCP is configured entirely via environment variables. Create a .env file or export variables directly:

| Variable | Description | Required | Default | |----------|-------------|----------|---------| | VISION_API_KEY | Your API key | Yes | - | | VISION_BASE_URL | API endpoint base URL | Yes | https://api.openai.com/v1 | | VISION_MODEL | Model name for vision | Yes | gpt-4o-mini | | VISION_MAX_TOKENS | Max response tokens | No | 4096 | | VISION_TIMEOUT | Request timeout (seconds) | No | 120 | | VISION_RESIZE | Resize image before analysis (WxH format, e.g., 1920x1080) | No | - |

Example .env File

VISION_API_KEY=your-api-key-here
VISION_BASE_URL=https://your-custom-endpoint.com/api/v1
VISION_MODEL=Qwen3.5-4B-AWQ
VISION_MAX_TOKENS=4096
VISION_TIMEOUT=120
VISION_RESIZE=1920x1080

Usage

Running the Server

# Using npx (recommended - always gets latest version)
npx -y @erickstryck/simple-vision-mcp

# Using global installation
simple-vision-mcp

# From source
npm start

# With environment variables inline
VISION_API_KEY=your-key VISION_BASE_URL=https://api.example.com/v1 VISION_MODEL=your-model npx -y @erickstryck/simple-vision-mcp

OpenCode Configuration

Add to your opencode.json:

{
  "mcp": {
    "vision": {
      "type": "local",
      "command": ["npx", "-y", "@erickstryck/simple-vision-mcp"],
      "env": {
        "VISION_API_KEY": "your-api-key",
        "VISION_BASE_URL": "https://your-endpoint.com/api/v1",
        "VISION_MODEL": "your-vision-model"
      },
      "enabled": true
    }
  }
}

Claude Desktop Configuration

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "vision": {
      "command": "npx",
      "args": ["-y", "@erickstryck/simple-vision-mcp"],
      "env": {
        "VISION_API_KEY": "your-api-key",
        "VISION_BASE_URL": "https://your-endpoint.com/api/v1",
        "VISION_MODEL": "your-vision-model"
      }
    }
  }
}

Cursor Configuration

Add to your Cursor MCP settings:

{
  "mcpServers": {
    "vision": {
      "command": "npx",
      "args": ["-y", "@erickstryck/simple-vision-mcp"],
      "env": {
        "VISION_API_KEY": "your-api-key",
        "VISION_BASE_URL": "https://your-endpoint.com/api/v1",
        "VISION_MODEL": "your-vision-model"
      }
    }
  }
}

Available Tools

analyze_image

Analyzes an image and returns a detailed description.

Parameters:

| Parameter | Type | Description | Required | |-----------|------|-------------|----------| | image_path | string | Path to the image file | Yes | | prompt | string | Custom analysis prompt | No | | width | number | Target width to resize the image before analysis | No | | height | number | Target height to resize the image before analysis | No |

Default Prompt: "Describe this image in detail, including objects, text, colors, composition, and any notable features."

Example:

{
  "name": "analyze_image",
  "arguments": {
    "image_path": "/path/to/image.png",
    "prompt": "What objects are in this image?"
  }
}

Response:

{
  "content": [
    {
      "type": "text",
      "text": "The image shows a red square with..."
    }
  ]
}

Supported Image Formats

• PNG (.png)
• JPEG (.jpg, .jpeg)
• GIF (.gif)
• WebP (.webp)
• BMP (.bmp)

Development

Project Structure

simple-vision-mcp/
├── src/
│   ├── config/
│   │   └── index.ts          # Configuration loading
│   ├── services/
│   │   └── visionService.ts  # Vision API client
│   ├── tools/
│   │   └── analyzeImage.ts   # MCP tool definition
│   ├── utils/
│   │   └── imageProcessor.ts # Image processing utilities
│   └── index.ts              # Main entry point
├── bin/
│   └── cli.js                # CLI wrapper
├── tests/
│   ├── config.test.ts
│   ├── imageProcessor.test.ts
│   └── visionService.test.ts
├── package.json
├── tsconfig.json
└── README.md

Building

npm run build

Testing

# Run tests once
npm test

# Watch mode
npm run test:watch

Design Principles

1. Single Responsibility: Each module has one clear purpose
2. Dependency Injection: Services receive dependencies via constructor
3. Functional Core: Business logic is pure and testable
4. Explicit over Implicit: Clear types and function signatures

Troubleshooting

"VISION_API_KEY environment variable is required"

Ensure you've set the VISION_API_KEY environment variable before starting the server.

"Unsupported image format"

The image format is not supported. Ensure your image is PNG, JPEG, GIF, WebP, or BMP format.

"Vision API error: 401"

Authentication failed. Verify your API key is correct and has access to vision capabilities.

"Vision API error: 4xx/5xx"

Check your VISION_BASE_URL is correct and the API endpoint is accessible.

License

MIT License - see LICENSE file for details.

Contributing

Contributions welcome! Please feel free to submit a Pull Request.