Text Classification MCP Server (Model2Vec)

A powerful Model Context Protocol (MCP) server that provides comprehensive text classification tools using fast static embeddings from Model2Vec (Minish Lab).

🛠️ Complete MCP Tools & Resources

This server provides 6 essential tools, 2 resources, and 1 prompt template for text classification:

🏷️ Classification Tools

• classify_text - Classify single text with confidence scores

• batch_classify - Classify multiple texts simultaneously

📝 Category Management Tools

• add_custom_category - Add individual custom categories

• batch_add_custom_categories - Add multiple categories at once

• list_categories - View all available categories

• remove_categories - Remove unwanted categories

📊 Resources

• categories://list - Access category list programmatically

• model://info - Get model and system information

💬 Prompt Templates

• classification_prompt - Ready-to-use classification prompt template

🚀 Key Features

• Multiple Transports: Supports stdio (local) and HTTP/SSE (remote) transports

• Fast Classification: Uses efficient static embeddings from Model2Vec

• 10 Default Categories: Technology, business, health, sports, entertainment, politics, science, education, travel, food

• Custom Categories: Add your own categories with descriptions

• Batch Processing: Classify multiple texts at once

• Resource Endpoints: Access category lists and model information

• Prompt Templates: Built-in prompts for classification tasks

• Production Ready: Docker, nginx, systemd support

📋 Installation

Prerequisites

• Python 3.10+

• uv package manager (recommended) or pip

Quick Setup

# Install dependencies
pip install -r requirements.txt

# Or with uv
uv sync 

🏃‍♂️ Running the Server

Option 1: Stdio Transport (Local/Traditional)

# Run with stdio (default - for Claude Desktop local config)
python text_classifier_server.py

# Or explicitly
python text_classifier_server.py --stdio

Option 2: HTTP Transport (Remote/Web)

# Run with HTTP transport on localhost:8000
python text_classifier_server.py --http

# Run on custom port
python text_classifier_server.py --http 9000

# Use the convenience script
./start_server.sh http 8000

Option 3: Using the HTTP Runner

# More options with the HTTP runner
python run_http_server.py --transport http --host 127.0.0.1 --port 8000 --debug

🔧 Configuration

For Claude Desktop

Stdio Transport (Local)

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

{
  "mcpServers": {
    "text-classifier": {
      "command": "python",
      "args": ["path/to/text_classifier_server.py"],
      "env": {}
    }
  }
}

HTTP Transport (Remote)

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

{
  "mcpServers": {
    "text-classifier-http": {
      "url": "http://localhost:8000/sse",
      "env": {}
    }
  }
}

For VS Code

Add to .vscode/mcp.json:

{
  "servers": {
    "text-classifier": {
      "type": "sse",
      "url": "http://localhost:8000/sse",
      "description": "Text classification server using static embeddings"
    }
  }
}

For Cursor IDE

Similar to Claude Desktop, but check Cursor's MCP documentation for the exact configuration path.

🛠️ Available Tools

classify_text

Classify a single text into predefined categories with confidence scores.

Parameters:

• text (string): The text to classify

• top_k (int, optional): Number of top categories to return (default: 3)

Returns: JSON with predictions, confidence scores, and category descriptions

Example:

classify_text("Apple announced new AI features", top_k=3)

batch_classify

Classify multiple texts simultaneously for efficient processing.

Parameters:

• texts (list): List of texts to classify

• top_k (int, optional): Number of top categories per text (default: 1)

Returns: JSON with batch classification results

Example:

batch_classify(["Tech news", "Sports update", "Business report"], top_k=2)

add_custom_category

Add a new custom category for classification.

Parameters:

• category_name (string): Name of the new category

• description (string): Description to generate the category embedding

Returns: JSON with operation result

Example:

add_custom_category("automotive", "Cars, vehicles, transportation, automotive industry")

batch_add_custom_categories

Add multiple custom categories in a single operation for efficiency.

Parameters:

• categories_data (list): List of dictionaries with 'name' and 'description' keys

Returns: JSON with batch operation results

Example:

batch_add_custom_categories([
    {"name": "automotive", "description": "Cars, vehicles, transportation"},
    {"name": "music", "description": "Music, songs, artists, albums, concerts"}
])

list_categories

List all available categories and their descriptions.

Parameters: None

Returns: JSON with all categories and their descriptions

remove_categories

Remove one or multiple categories from the classification system.

Parameters:

• category_names (list): List of category names to remove

Returns: JSON with removal results for each category

Example:

remove_categories(["automotive", "custom_category"])

📚 Available Resources

• categories://list: Get list of available categories with metadata

• model://info: Get information about the loaded Model2Vec model and system status

💬 Available Prompts

• classification_prompt: Template for text classification tasks with context and instructions

Parameters:

• text (string): The text to classify

Returns: Formatted prompt for classification with available categories listed

🧪 Testing

Test HTTP Server

# Test the HTTP server endpoints
python test_http_client.py

# Check server status
./check_server.sh

# Test with curl
curl http://localhost:8000/sse

Test with MCP Inspector

# For stdio transport
mcp dev text_classifier_server.py

# For HTTP transport (start server first)
# Then connect MCP Inspector to http://localhost:8000/sse

🐳 Docker Deployment

Basic Docker

# Build and run
docker build -t text-classifier-mcp .
docker run -p 8000:8000 text-classifier-mcp

Docker Compose

# Basic deployment
docker-compose up

# With nginx reverse proxy
docker-compose --profile production up

🚀 Production Deployment

Systemd Service

# Copy service file
sudo cp text-classifier-mcp.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl enable text-classifier-mcp
sudo systemctl start text-classifier-mcp

Nginx Reverse Proxy

The included nginx.conf provides:

• HTTP/HTTPS termination

• Proper SSE headers

• Load balancing support

• SSL configuration template

🌐 Transport Comparison

🔍 Troubleshooting

Common Issues

1. Server won't start

   # Check if port is in use
   lsof -i :8000
   
   # Try different port
   python run_http_server.py --port 9000

2. Claude Desktop connection fails

   # Check server status
   ./check_server.sh
   
   # Verify config file syntax
   cat ~/Library/Application\ Support/Claude/claude_desktop_config.json | python -m json.tool

3. Model download fails

   # Manual model download
   python -c "from model2vec import StaticModel; StaticModel.from_pretrained('minishlab/potion-base-8M')"

Debug Mode

# Enable debug logging
python run_http_server.py --debug

# Check logs
tail -f logs/mcp_server.log

📖 Technical Details

• Model: minishlab/potion-base-8M from Model2Vec

• Similarity: Cosine similarity between text and category embeddings

• Performance: ~30MB model, fast inference with static embeddings

• Protocol: MCP specification 2024-11-05

• Transports: stdio, HTTP+SSE, Streamable HTTP

🤝 Contributing

1. Fork the repository

2. Create a feature branch

3. Add tests for new functionality

4. Submit a pull request

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• Model2Vec by Minish Lab for fast static embeddings

• Anthropic for the Model Context Protocol specification

• FastMCP for the excellent Python MCP framework

Need help? Check the troubleshooting section or open an issue in the repository.