MCP SOP Server

A Model Context Protocol (MCP) server for accessing and searching Standard Operating Procedures (SOPs) with Italian language support.

Overview

This MCP server provides AI agents with the ability to:

• Search through your company's SOP documentation using semantic search
• Retrieve relevant procedures based on specific situations
• Browse SOPs by category
• Get guidance on what to do in specific scenarios

The server uses:

• ChromaDB for vector storage and semantic search
• Sentence Transformers with multilingual models for Italian language support
• FastMCP for the MCP server implementation
• RAG (Retrieval Augmented Generation) for intelligent document retrieval

Features

• 🇮🇹 Italian Language Support: Uses multilingual embeddings optimized for Italian text
• 📄 Multi-format Support: Processes PDF and DOCX documents
• 🔍 Semantic Search: Find relevant SOPs based on meaning, not just keywords
• 📁 Category Filtering: Search within specific SOP categories
• 🤖 AI-Ready: Provides structured responses perfect for LLM consumption
• ⚡ Fast Retrieval: Efficient vector-based search with ChromaDB
• 🚀 Lazy Initialization: Server starts quickly, documents are indexed on first request

Example Queries

Here are some example queries you can use:

Italian:

• "Come gestire una non conformità?"
• "Procedura per la manutenzione dell'infrastruttura"
• "Cosa fare in caso di audit interno?"
• "Gestione del magazzino e inventario"

English:

• "How to handle quality issues?"
• "Software development lifecycle procedures"
• "Risk management protocols"
• "Employee training requirements"

Development

Project Structure

mcp-sop-server/
├── src/mcp_sop_server/          # Main package
│   ├── __init__.py              # Package initialization
│   ├── mcp_server.py            # FastMCP server and tools
│   ├── document_processor.py    # Document text extraction
│   └── document_searcher.py     # Vector search with ChromaDB
├── main.py                      # Entry point
├── requirements.txt             # Dependencies
├── test_server.py              # Server testing
├── mcp-client-config.json      # Example client configuration
└── README.md                   # This file

Adding New Document Types

To support additional file formats, extend the DocumentProcessor class:

def extract_text_from_new_format(self, file_path: Path) -> str:
    # Implementation for new format
    pass

Custom Search Logic

Modify the DocumentSearcher class to implement custom search algorithms or filters.

Additional Tools

Add new MCP tools by defining them with the @mcp.tool() decorator in mcp_server.py.

Support

For issues or questions about the SOP MCP server, create an issue on the GitHub repository.

Installation

1. Clone the repository:

   Copy

   git clone https://github.com/dadapera/mcp-sop-server.git
   cd mcp-sop-server

2. Create and activate virtual environment:

   Copy

   python -m venv venv
   
   # On Windows
   venv\Scripts\activate
   
   # On macOS/Linux
   source venv/bin/activate

3. Install Dependencies:

   Copy

   pip install -r requirements.txt

4. Add your SOP documents: Create a sop_documents/ directory and organize your SOP files by category folders.

Configuration

MCP Client Setup

To use this server with Claude Desktop or other MCP clients, add it to your MCP client configuration:

For Claude Desktop (mcp-client-config.json):

{
  "mcpServers": {
    "sop-server": {
      "command": "/path/to/your/venv/Scripts/python.exe",
      "args": ["/path/to/your/mcp-sop-server/main.py"],
      "cwd": "/path/to/your/mcp-sop-server"
    }
  }
}

Note: Make sure to use the full path to your virtual environment's Python executable and adjust paths according to your system.

Directory Structure

The server expects SOP documents to be organized as follows:

sop_documents/
├── SOP01 Quality System Documentation Management/
│   ├── document1.pdf
│   └── document2.docx
├── SOP02 HR management/
│   └── hr_procedures.pdf
├── SOP03 Design/
│   └── design_process.docx
└── ...

Usage

Starting the Server

The server is typically started automatically by your MCP client (like Claude Desktop). If running manually:

python main.py

The server will:

1. Start quickly and wait for connections
2. On first tool call: scan all SOP documents in the sop_documents/ directory
3. Process and extract text from PDF and DOCX files
4. Generate embeddings using the multilingual model
5. Store everything in ChromaDB for fast retrieval

Available Tools

The server provides the following MCP tools:

1. search_sop_documents

Search for relevant SOP content using natural language queries.

Parameters:

• query (string): Search query in Italian or English
• max_results (int, optional): Maximum results to return (default: 5)
• category (string, optional): Filter by SOP category

Example:

{
  "query": "Come gestire una non conformità nel processo di produzione",
  "max_results": 3,
  "category": "SOP05 Non Conformity"
}

2. get_sop_guidance

Get specific guidance for a situation based on SOP documents.

Parameters:

• situation (string): Description of the situation
• category (string, optional): Focus search on specific category

Example:

{
  "situation": "Un cliente ha segnalato un difetto nel prodotto consegnato",
  "category": "SOP05 Non Conformity"
}

3. list_sop_categories

Get all available SOP categories and collection statistics.

4. get_sop_by_category

Retrieve all SOPs within a specific category.

Parameters:

• category (string): SOP category name

5. refresh_sop_database

Refresh the document database (use when SOPs are updated).

6. get_server_status

Get current server status and statistics.

Technical Configuration

Embedding Model

The server uses sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 by default for Italian language support. You can change this in src/mcp_sop_server/document_searcher.py:

model_name = "sentence-transformers/your-preferred-model"

Chunk Size

Text chunking can be adjusted in src/mcp_sop_server/document_processor.py:

def chunk_text(self, text: str, chunk_size: int = 1000, overlap: int = 200):

Database Location

ChromaDB storage location is automatically set to chroma_db/ in the project root.

Troubleshooting

Common Issues

1. No documents found: Ensure SOP documents are in the correct directory structure
2. Embedding model download: First run may take time to download the multilingual model
3. Memory usage: Large document collections may require more RAM for embedding generation
4. Path issues: Make sure to use absolute paths in your MCP client configuration

Logs

The server provides detailed logging with emojis for better readability:

• 🚀 Server startup and initialization
• 📄 Document processing status
• 📊 Processing statistics
• ✅ Success messages
• ❌ Error messages

Check the console output or your MCP client logs for detailed information.

License

This project is intended for internal company use for accessing SOP documentation.