ron-gabriel/mcp-server

MCP Email Processing Server

A Model Context Protocol (MCP) server that provides tools for LLMs to access and process emails from Microsoft 365 Outlook for various business workflows.

Security Notice: This application handles sensitive data. Please review before deployment.

Architecture

The workflow follows this pattern:

1. Email arrives → Email arrives in designated inbox
2. MCP server reads email → Provides raw email data to LLM
3. LLM processes data → Analyzes content, extracts information
4. LLM uses MCP tools → Sends formatted data to your API

The MCP server acts as a bridge between the email system and the LLM, providing simple data access tools while the LLM handles all the intelligence and decision-making.

Setup

1. Azure AD App Registration

   • Register an app in Azure AD
   • Add Microsoft Graph API permissions:
     • Mail.Read
     • Mail.ReadWrite
   • Create a client secret

2. Configuration

   • Copy .env.example to .env
   • Fill in your Azure AD credentials:

     AZURE_CLIENT_ID=your-client-id
     AZURE_CLIENT_SECRET=your-client-secret
     AZURE_TENANT_ID=your-tenant-id
     OUTLOOK_USER_EMAIL=processing@yourdomain.com
     OUTLOOK_FOLDER_NAME=Inbox
     API_ENDPOINT=https://your-api.com/webhook
     API_KEY=your-api-key
     

3. Installation

   npm install
   npm run build
   npm start
   

Available Tools

Email Reading Tools

• get_unread_emails - Retrieves a list of unread emails with basic metadata
• get_email_content - Gets the full content of a specific email (HTML or plain text)
• search_emails - Searches for emails containing specific keywords
• get_email_conversation - Retrieves all emails in a conversation thread

Processing Tools

• send_to_api - Sends LLM-formatted data to your API endpoint
• mark_email_processed - Marks an email as read/processed

Example LLM Workflow

1. LLM calls get_unread_emails to see new messages
2. For each email:
   - LLM calls get_email_content to read the full message
   - LLM analyzes the content for:
     - Sentiment and tone
     - Intent and purpose
     - Key information (dates, references, entities)
   - LLM formats the analysis into structured data
   - LLM calls send_to_api with the formatted data
   - LLM calls mark_email_processed to mark as complete

API Data Format

The LLM can send any JSON structure to your API. A typical format might be:

{
  "emailId": "AAMkAGI2...",
  "from": "sender@example.com",
  "receivedAt": "2024-01-10T10:30:00Z",
  "analysis": {
    "category": "inquiry",
    "priority": "high",
    "confidence": 0.95
  },
  "extractedData": {
    "subject": "Product inquiry",
    "entities": ["Product X", "Feature Y"],
    "requestType": "information",
    "referenceNumber": "REF-1234"
  },
  "summary": "Customer requesting information about Product X features..."
}

Benefits of this Architecture

• Separation of Concerns: MCP server handles email access, LLM handles intelligence
• Flexibility: LLM can adapt to any email format or analysis requirement
• No Hardcoded Logic: Analysis patterns aren't fixed in code
• Easy Updates: Change analysis behavior by updating LLM prompts, not code
• Rich Context: LLM can consider conversation history and nuanced language

Running the Server

The server runs on stdio and connects to MCP-compatible LLM clients:

npm start

Documentation

• - Complete deployment instructions
• - Security best practices and considerations
• - Day-to-day operations and monitoring

Quick Start

1. Clone the repository
2. Copy .env.example to .env and configure
3. Run npm install and npm run build
4. Deploy using Docker or Kubernetes (see )

Support

For issues and questions:

1. Check the troubleshooting section
2. Review logs using kubectl logs or docker logs
3. File an issue with sanitized logs and configuration