Kaltura MCP Server

# Retention Analysis Prompt Example This document demonstrates how an LLM (like Claude) would use the retention analysis prompt in the Kaltura MCP server. ## How the LLM Accesses the Prompt ### 1. Discovery Phase When the LLM connects to the Kaltura MCP server, it can discover available prompts by calling the `list_prompts` function. The server will return: ```json [ { "name": "retention_analysis", "description": "Create comprehensive retention analysis report with visualizations and recommendations", "arguments": [ { "name": "entry_id", "description": "Video entry ID to analyze (e.g., '1_3atosphg')", "required": true }, { "name": "time_period", "description": "Number of months of data to analyze (default: 12)", "required": false }, { "name": "output_format", "description": "Report format: 'interactive' (HTML) or 'markdown' (default: interactive)", "required": false } ] } // ... other prompts ] ``` ### 2. Invocation Phase When a user asks for a retention analysis, the LLM would invoke the prompt: ```json { "method": "get_prompt", "params": { "name": "retention_analysis", "arguments": { "entry_id": "1_3atosphg", "time_period": "6", "output_format": "interactive" } } } ``` ### 3. Execution Phase The server returns a structured conversation flow that guides the LLM through the analysis: ```json { "messages": [ { "role": "user", "content": "Create an interactive retention-analysis report for video **1_3atosphg** using the last 6 months of viewer data." }, { "role": "assistant", "content": "I'll create a comprehensive retention analysis report with visualizations and recommendations. Let me gather all the necessary data." }, { "role": "user", "content": "First, use get_video_retention with entry_id='1_3atosphg', from_date='2024-07-01', to_date='2025-01-01' to get the retention curve data. Then use get_media_entry with entry_id='1_3atosphg' to get video metadata including duration." }, { "role": "user", "content": "Next, check for available content metadata:\n1. Use list_caption_assets with entry_id='1_3atosphg' to find transcripts\n2. If captions exist, use get_caption_content to retrieve the transcript\n3. Look for cue points and chapter markers in the video metadata" }, { "role": "user", "content": "Now analyze the retention data to create:\n1. **Retention curve** - Convert percentiles to timestamps (hh:mm:ss format)\n2. **Drop-off table** - Identify 10-second windows with ≥5% retention loss\n3. **Content insights** - Correlate drop-offs with transcript/metadata\n4. **Recommendations** - Suggest edits, CTAs, and highlight clips" }, { "role": "user", "content": "Format the output as an interactive HTML report with:\n- Hoverable retention curve showing insights in tooltips\n- Sortable drop-off table\n- Expandable recommendation sections\n- Embedded video player with timestamp links" } ] } ``` ## How the LLM Follows the Workflow The LLM would then execute the suggested tools in sequence: ### Step 1: Get Retention Data ```python # LLM calls get_video_retention result = await get_video_retention( entry_id="1_3atosphg", from_date="2024-07-01", to_date="2025-01-01" ) ``` ### Step 2: Get Video Metadata ```python # LLM calls get_media_entry metadata = await get_media_entry(entry_id="1_3atosphg") # Extract duration to convert percentiles to timestamps ``` ### Step 3: Check for Captions ```python # LLM calls list_caption_assets captions = await list_caption_assets(entry_id="1_3atosphg") if captions: # Get transcript for content correlation transcript = await get_caption_content(caption_asset_id=captions[0].id) ``` ### Step 4: Generate Report The LLM would then: 1. Convert retention percentiles (0-100) to video timestamps using the duration 2. Identify drop-off points where retention decreases by ≥5% 3. Correlate drop-offs with transcript content 4. Generate recommendations based on patterns 5. Format everything as an interactive HTML report ## Example User Interaction **User**: "Analyze the retention for video 1_3atosphg over the last 6 months and create an interactive report" **Assistant**: "I'll analyze the retention data for video 1_3atosphg and create a comprehensive interactive report. Let me gather the data..." [Assistant executes the workflow from the prompt] **Assistant**: "Here's your retention analysis report: ### Video Retention Analysis Report **Video ID**: 1_3atosphg **Analysis Period**: Last 6 months **Total Duration**: 12:34 #### Key Findings: - Average retention: 68.5% - Major drop-off at 02:15 (-8.3%) - Highest engagement: 08:45-09:30 - 45% completion rate [Interactive HTML visualization would be embedded here] #### Recommendations: 1. **Edit Suggestion**: Consider trimming content at 02:10-02:30 where 8.3% of viewers drop off 2. **Highlight Reel**: Create a 2-minute clip from 08:45-10:45 (highest engagement) 3. **CTA Placement**: Add call-to-action at 11:30 before final drop-off Would you like me to generate the highlight reel timestamps or analyze a different video?" ## Benefits of This Approach 1. **Structured Workflow**: The prompt ensures the LLM follows best practices 2. **Complete Analysis**: All necessary data is gathered systematically 3. **Flexible Output**: Supports both interactive and markdown formats 4. **Actionable Insights**: Delivers specific recommendations, not just data 5. **Reusable Pattern**: Users can easily analyze multiple videos ## Integration with MCP This prompt template leverages MCP's prompt system to: - **Reduce Complexity**: Users don't need to know the exact tool sequence - **Ensure Consistency**: Every retention analysis follows the same thorough process - **Enable Discovery**: LLMs can suggest this analysis when appropriate - **Improve Results**: Structured workflow leads to better, more complete reports