eagleisbatman/agrivision-mcp-server

AgriVision MCP Server

AI-Powered Plant Disease & Health Diagnosis via Model Context Protocol

AgriVision is a region-agnostic plant health diagnosis service providing structured diagnostic data through MCP. Uses Google Gemini 2.5 Flash for accurate crop disease detection, pest identification, and nutrient deficiency analysis.

🌟 Features

• AI Vision Analysis: Powered by Google Gemini 2.5 Flash (configurable)
• 23 Supported Crops: Cereals, legumes, vegetables, cash crops
• Disease Detection: Fungal, bacterial, viral with scientific names
• Pest Identification: Insects, mites, infestations
• Nutrient Analysis: Deficiency detection
• Growth Stage Assessment: Seedling to senescent
• Configurable Modes: Diagnosis-only or full advisory
• Structured Output: JSON or formatted text
• Region-Agnostic: Adaptable to any geography

🤖 AI Model: Google Gemini 2.5 Flash

Alternative Models: gemini-2.0-flash-exp, gemini-1.5-pro, gemini-1.5-flash Configure via GEMINI_IMAGE_MODEL environment variable.

🚀 Quick Start

git clone https://github.com/eagleisbatman/agrivision-mcp-server.git
cd agrivision-mcp-server
npm install
cp .env.example .env
# Add your GEMINI_API_KEY to .env
npm run dev

Get API Key: https://makersuite.google.com/app/apikey

⚙️ Configuration

Required

GEMINI_API_KEY=your_key_here

Advisory Mode (Important!)

ADVISORY_MODE=diagnosis_only  # or "full_advisory"

• diagnosis_only (default): Returns diagnostic data only. Client agent generates region-specific treatment.
• full_advisory: Returns diagnosis + AI treatment recommendations.

Output Format

OUTPUT_FORMAT=structured  # or "text"

• structured (default): JSON with typed fields
• text: Formatted text with emojis

Model Selection

GEMINI_IMAGE_MODEL=gemini-2.5-flash-image-preview

🔌 API Endpoints

GET /health

{
  "status": "healthy",
  "model": "gemini-2.5-flash-image-preview",
  "advisoryMode": "diagnosis_only",
  "supportedCrops": 23
}

POST /mcp

MCP protocol endpoint for tool calls.

🛠️ MCP Tool: diagnose_plant_health

Parameters

• image (required): Base64-encoded image (data:image/jpeg;base64,...)
• crop (optional): Expected crop type (e.g., maize, tomato)

Supported Crops (23)

Cereals: maize, wheat, rice, sorghum, millet Legumes: beans, cowpea, pigeon_pea, groundnut Roots: cassava, sweet_potato, potato Vegetables: tomato, cabbage, kale, onion Cash Crops: tea, coffee, sugarcane, banana, sunflower, cotton

Response (diagnosis_only + structured)

{
  "crop": {"name": "Maize", "scientific_name": "Zea mays", "confidence": "High"},
  "health_status": {"overall": "Moderate Issue", "confidence": "High"},
  "issues": [{
    "name": "Fall Armyworm",
    "scientific_name": "Spodoptera frugiperda",
    "category": "Pest",
    "severity": "Moderate",
    "symptoms": ["Irregular holes", "Frass visible"]
  }],
  "growth_stage": "Vegetative"
}

🌍 Region-Agnostic Design

Diagnostic data uses scientific terminology. Client agents add regional context.

Architecture:

Image → AgriVision (diagnosis_only) → Agent Builder → Region-specific advice

Example: Same diagnosis, different regions:

• Kenya: "Use Neem oil from local agro-dealer"
• India: "Apply Bt from Krishi Bhawan"
• USA: "Use recommended Bt from ag store"

📦 Deploy to Railway

1. Push to GitHub
2. Create project on railway.app from GitHub repo
3. Set environment variables:
   • GEMINI_API_KEY
   • ADVISORY_MODE=diagnosis_only
   • OUTPUT_FORMAT=structured
4. Deploy automatically
5. Test: curl https://your-app.up.railway.app/health

🔗 OpenAI Agent Builder Integration

1. Add MCP connection:

   • URL: https://your-app.up.railway.app/mcp
   • Type: StreamableHTTP

2. System Prompt:

When users share images, use `diagnose_plant_disease` tool.
Tool returns diagnostic data only.
YOU must generate region-specific treatment advice.

🧪 Testing

node test-diagnosis.js /path/to/image.jpg

📄 License

MIT - Built with ❤️ by Digital Green Foundation

Support: GitHub Issues or support@digitalgreen.org