README - Kaltura Model Context Protocol (MCP) Server by zoharbabin

Kaltura MCP Server

A Model Context Protocol (MCP) server that provides secure, read-only tools for managing Kaltura API operations. This server enables AI assistants to search, discover, and analyze Kaltura media content safely.

Features

Media Discovery: Search and browse media entries with advanced filtering
Content Analysis: Access captions, transcripts, and attachment content
Category Management: Browse and explore content categories
Analytics: Retrieve viewing analytics and performance metrics
Secure Access: Read-only operations with comprehensive input validation
Session Management: Automatic session handling with configurable expiry

Installation

Clone this repository:

git clone https://github.com/zoharbabin/kaltura-mcp.git
cd kaltura-mcp

Install dependencies:

pip install -e .

Usage Modes

This server supports two deployment modes:

🔧 Local MCP Server (Stdio Mode)

Best for: Personal use, direct Claude Desktop integration, development

🌐 Remote MCP Server (HTTP/SSE Mode)

Best for: Hosted services, multiple users, production deployments

Local MCP Server Setup (Claude Desktop)

Step 1: Install Package

pip install kaltura-mcp

Step 2: Setup Environment Configuration

🔒 Secure Method (Recommended): Use the interactive setup script:

# Navigate to your project directory
cd /path/to/kaltura-mcp

# Run the interactive setup
python setup_env.py

The script will guide you through:

Choosing between stdio (local) or remote mode
Securely entering your Kaltura credentials
Generating a .env file with proper permissions (600)
Providing the exact Claude Desktop configuration

📋 Manual Method: Copy and edit the example file:

# Copy the example file
cp .env.example .env

# Edit with your credentials
# - For stdio mode: Only fill in KALTURA_* variables
# - For remote mode: Fill in JWT_SECRET_KEY, OAUTH_*, and SERVER_* variables
nano .env

# Set secure permissions
chmod 600 .env

Step 3: Get Your Kaltura Credentials

You'll need these credentials from your Kaltura account:

Service URL: Your Kaltura server URL (usually https://cdnapisec.kaltura.com)
Partner ID: Your numeric partner ID (found in KMC → Settings → Integration Settings)
Admin Secret: Your API admin secret key (found in KMC → Settings → Integration Settings)
User ID: Your Kaltura user ID (usually your email or admin)

Step 4: Configure Claude Desktop

Open your Claude Desktop configuration file:

macOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

🔒 Secure Configuration (credentials in .env file):

{
  "mcpServers": {
    "kaltura": {
      "command": "/full/path/to/kaltura-mcp"
    }
  }
}

Important Notes:

Replace /full/path/to/kaltura-mcp with the actual path to your kaltura-mcp command (find it with which kaltura-mcp)
The .env file is automatically loaded from the project directory by the server
The setup_env.py script will automatically detect and provide the correct command path

Step 5: Restart Claude Desktop

After saving the configuration file, restart Claude Desktop completely for the changes to take effect.

Step 6: Test the Integration

In Claude Desktop, try asking:

"Search for recent Kaltura videos"
"List my Kaltura categories"
"Find videos about [topic] in my Kaltura account"

Troubleshooting Local Setup

Issue: kaltura-mcp command not found

Solution: Make sure you installed with pip install kaltura-mcp and the command is in your PATH

Issue: "Error: Missing required environment variables"

Solution:
1. Check that .env file exists in your project directory
2. Verify file permissions: ls -la .env (should show -rw-------)
3. Ensure all required Kaltura credentials are set in .env file

Issue: "Invalid credentials" or "Authentication failed"

Solution:
1. Verify your credentials in Kaltura KMC → Settings → Integration Settings
2. Check .env file for typos or extra spaces
3. Run python setup_env.py to recreate the configuration

Issue: Claude Desktop doesn't show the MCP server

Solution:
1. Check the config file syntax with a JSON validator
2. Verify the command path is correct (use which kaltura-mcp)
3. Restart Claude Desktop completely
4. Check Claude Desktop logs for error messages

Issue: ".env file not found" error

Solution:
1. Run python setup_env.py from your project directory
2. Ensure the .env file exists in the same directory as the server code
3. Check file permissions: ls -la .env (should show -rw-------)

✅ Security Benefits:

✅ Secure file permissions (600 - owner only)
✅ Git-ignored by default (won't be committed)
✅ Local to project directory (easy to manage)
✅ Standard .env pattern (familiar to developers)
✅ No credentials in config files (improved security)

Remote MCP Server Setup

Configuration

For remote/hosted deployment, additional environment variables are required:

cp .env.example .env
# Configure remote server settings

Required environment variables:

JWT_SECRET_KEY: Strong secret key for JWT token signing (⚠️ CRITICAL FOR SECURITY)
OAUTH_REDIRECT_URI: OAuth callback URL (e.g., https://your-domain.com/oauth/callback)
SERVER_HOST: Server bind address (default: 0.0.0.0)
SERVER_PORT: Server port (default: 8000)

Optional environment variables:

SERVER_RELOAD: Enable auto-reload in development (default: false)
OAUTH_CLIENT_ID: Custom OAuth client ID
OAUTH_CLIENT_SECRET: Custom OAuth client secret

Running Remote Server

# Using the installed command
kaltura-mcp-remote

# Or using Python module
python -m kaltura_mcp.remote_server

The remote server provides:

HTTP/SSE transport for MCP protocol
JWT-based authentication for secure credential management
Web-based authorization flow for user-friendly setup
Multi-tenant support for hosting as a service

Available Tools

get_media_entry - Get detailed information about a specific media entry
- Parameters: entry_id (required)
list_categories - List and search content categories
- Parameters: search_text, limit
Analytics Tools - Comprehensive analytics suite with purpose-driven functions:
- get_analytics - General analytics data for reporting and analysis
- get_analytics_timeseries - Time-series data optimized for charts
- get_video_retention - Detailed viewer retention analysis throughout videos
- get_realtime_metrics - Live analytics updated every ~30 seconds
- get_quality_metrics - Quality of Experience (QoE) and streaming performance
- get_geographic_breakdown - Location-based analytics at various granularities
- list_analytics_capabilities - Discover all available analytics functions
- See for detailed usage
get_download_url - Get direct download URL for media files
- Parameters: entry_id (required), flavor_id
get_thumbnail_url - Get video thumbnail/preview image URL with custom dimensions
- Parameters: entry_id (required), width, height, second
search_entries - Search and discover media entries with intelligent sorting and filtering
- Parameters: query (required), search_type, match_type, specific_field, boolean_operator, include_highlights, custom_metadata, date_range, max_results, sort_field, sort_order
list_caption_assets - List available captions and subtitles for a media entry
- Parameters: entry_id (required)
get_caption_content - Get caption/subtitle content and download URL
- Parameters: caption_asset_id (required)
list_attachment_assets - List attachment assets for a media entry
- Parameters: entry_id (required)
get_attachment_content - Get attachment content details and download content as base64
- Parameters: attachment_asset_id (required)

Prompts

The server provides intelligent prompts to guide users through complex workflows:

analytics_wizard - Interactive guide for creating comprehensive analytics reports

Arguments:
- analysis_goal: What to analyze (e.g., "video performance", "viewer engagement", "geographic reach")
- time_period: Time range (e.g., "today", "yesterday", "last_week", "last_month")

content_discovery - Natural language search assistant for finding media

Arguments:
- search_intent: What you're looking for in natural language
- include_details: Whether to fetch captions/attachments (yes/no)

accessibility_audit - Content accessibility compliance checker

Arguments:
- audit_scope: What to audit ("all", "recent", "category:name", or entry_id)

retention_analysis - Create comprehensive retention analysis report

Arguments:
- entry_id: Video to analyze (e.g., "1_3atosphg") [required]
- time_period: Months of data to analyze (default: "12")
- output_format: "interactive" (HTML) or "markdown" (default: "interactive")

Resources

The server exposes frequently-used data as cached resources:

kaltura://analytics/capabilities - Complete analytics documentation
- All 60+ report types with descriptions
- Available metrics and dimensions
- Best practices for different use cases
- Cached for 30 minutes
kaltura://categories/tree - Category hierarchy with entry counts
- Complete category tree structure
- Entry counts per category
- Parent-child relationships
- Cached for 30 minutes
kaltura://media/recent/{count} - Recent media entries
- Replace {count} with number of entries (e.g., kaltura://media/recent/20)
- Maximum 100 entries
- Includes basic metadata
- Cached for 5 minutes

Remote MCP Server (Advanced)

User Authorization Flow

Server Deployment: Deploy the remote server to your hosting environment
User Authorization: Users visit https://your-server.com/oauth/authorize
Credential Entry: Users securely enter their Kaltura credentials via web form
Token Generation: Server generates a JWT token with encrypted credentials
Client Configuration: Users add the server URL and token to their MCP client

Step-by-Step Remote Setup

1. Generate Secure JWT Secret

# Generate a strong secret key
python -c "import secrets; print(secrets.token_urlsafe(32))"

2. Configure Environment

# Set in your .env file or environment
JWT_SECRET_KEY=your-generated-secret-key-here
OAUTH_REDIRECT_URI=https://your-domain.com/oauth/callback
SERVER_HOST=0.0.0.0
SERVER_PORT=8000

3. Deploy Server

Option A: Direct Python

kaltura-mcp-remote

Option B: Docker

docker-compose up -d

Option C: Production with Gunicorn (Optional)

# Install gunicorn separately if needed for production
pip install gunicorn
gunicorn -w 4 -k uvicorn.workers.UvicornWorker kaltura_mcp.remote_server:app

4. User Onboarding

Send users to: https://your-server.com/oauth/authorize?response_type=code&client_id=kaltura-mcp&redirect_uri=https://your-server.com/oauth/callback&state=user123

5. Client Configuration

For Claude Desktop (Remote Mode):

The easiest way to use the remote server with Claude Desktop is via the proxy client:

{
  "mcpServers": {
    "kaltura-remote": {
      "command": "kaltura-mcp-proxy",
      "env": {
        "KALTURA_REMOTE_SERVER_URL": "https://your-server.com/mcp/messages",
        "KALTURA_REMOTE_ACCESS_TOKEN": "your-jwt-token-from-authorization-flow"
      }
    }
  }
}

The proxy client (kaltura-mcp-proxy) acts as a local stdio MCP server that forwards requests to your remote server. This provides the best compatibility with Claude Desktop.

For Custom MCP Clients:

// HTTP transport with authentication
const transport = new HTTPTransport({
  baseUrl: "https://your-server.com/mcp/messages",
  headers: {
    "Authorization": "Bearer user-jwt-token-here"
  }
});

Analytics Documentation

The MCP server provides a comprehensive analytics suite with purpose-driven functions optimized for different use cases:

Purpose-Built Analytics Functions:

get_analytics: Comprehensive reporting data in table format for detailed analysis
get_analytics_timeseries: Time-series data optimized for charts and visualizations
get_video_retention: Detailed viewer retention curves showing exactly where viewers drop off
get_realtime_metrics: Live analytics updated every ~30 seconds for monitoring
get_quality_metrics: Quality of Experience (QoE) metrics for streaming performance
get_geographic_breakdown: Location-based analytics at country, region, or city level

Analytics Capabilities:

60+ report types covering content, users, geography, platforms, and more
Raw data access for custom analysis and visualization
Intelligent insights including drop-off points and engagement patterns
Support for filtering by date ranges, categories, users, and dimensions

For comprehensive documentation, see:

- Complete reference for all analytics functions
- Code examples and visualizations

Security Considerations

Production Deployment

Use HTTPS: Always deploy with TLS/SSL certificates
Secure JWT Secret: Use a cryptographically strong secret key (32+ bytes)
Environment Security: Never commit secrets to version control
Network Security: Use firewalls and VPN access where appropriate
Regular Updates: Keep dependencies updated for security patches

JWT Token Security

Token Expiration: Tokens expire after 24 hours by default
Credential Encryption: Kaltura credentials are encrypted within JWT payload
Scope Limitation: Tokens are limited to read-only Kaltura operations
Revocation: Restart server to invalidate all existing tokens

Infrastructure

# Example nginx configuration for production
server {
    listen 443 ssl;
    server_name your-kaltura-mcp.com;
    
    ssl_certificate /path/to/cert.pem;
    ssl_certificate_key /path/to/key.pem;
    
    location / {
        proxy_pass http://localhost:8000;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }
}

Docker Deployment

docker-compose.yml for production:

version: '3.8'
services:
  kaltura-mcp:
    build: .
    ports:
      - "8000:8000"
    environment:
      - JWT_SECRET_KEY=${JWT_SECRET_KEY}
      - OAUTH_REDIRECT_URI=https://your-domain.com/oauth/callback
      - SERVER_HOST=0.0.0.0
      - SERVER_PORT=8000
    restart: unless-stopped
    volumes:
      - ./logs:/app/logs
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.kaltura-mcp.rule=Host(\`your-domain.com\`)"
      - "traefik.http.routers.kaltura-mcp.tls=true"

Monitoring & Logging

The remote server provides built-in logging and can be monitored via:

Health Check: GET / returns server status
Metrics: Access logs via Docker volumes or server logs
Error Tracking: Configure external error tracking services

Important Security Notes

Local Mode Security (Recommended)

✅ Direct Configuration - Credentials configured directly in Claude Desktop
✅ MCP Standard Compliance - Client passes credentials to server via environment variables
✅ Process Isolation - MCP server runs in isolated process with limited scope
✅ No network exposure - Direct API communication with Kaltura
✅ Local credential storage - Credentials never leave your machine
✅ Secure transmission - Credentials passed securely to MCP server process

Remote Mode Security

✅ Credential encryption - Kaltura credentials encrypted in JWT tokens
✅ Token expiration - Automatic 24-hour token expiry
✅ TLS encryption - HTTPS required for production
⚠️ Server trust - You must trust the remote server operator
⚠️ Credential transmission - Credentials are sent to remote server (encrypted)

Production Checklist

Use HTTPS with valid certificates
Generate strong JWT secret key (32+ bytes)
Configure secure environment variables
Set up proper logging and monitoring
Implement rate limiting (nginx/cloudflare)
Regular security updates
Backup and disaster recovery plan

Deployment Architectures

Personal Use (Recommended)

Claude Desktop ←→ Local MCP Server ←→ Kaltura API

Small Team

Claude Desktop ←→ Proxy Client ←→ Remote MCP Server ←→ Kaltura API

Enterprise

Multiple Clients ←→ Load Balancer ←→ Multiple MCP Servers ←→ Kaltura API
                                          ↓
                                    Redis/Database

Documentation

- Comprehensive guide to analytics features
- Detailed documentation of MCP prompts and resources
API Documentation - Official Kaltura API documentation

Development

Running Tests

pytest

Code Formatting

black src/
ruff check src/

License

MIT