Embed

Video Embeddings

Create video embeddings, check embedding task status, and retrieve embeddings from tasks or indexed videos using TwelveLabs.

When to Use This Skill

Use this skill when the user:

• Wants to create embeddings from a video ("embed this video", "create embeddings for this video")
• Asks about embedding task status ("are my embeddings ready?", "check embedding status")
• Wants to retrieve embeddings from an indexed video ("get embeddings for video abc123")
• Mentions video embeddings in any context

Instructions

Creating Embeddings from a Video

Step 1: Identify the Video Source

Extract the video path or URL from the user's message:

• Local file path: /path/to/video.mp4, ./video.mp4, ~/Videos/demo.mov
• Remote URL: https://example.com/video.mp4

If no path/URL is provided, ask the user:

Please provide the video file path or URL you'd like to create embeddings for.

Examples:
- Local file: /path/to/video.mp4 or ./video.mp4
- Remote URL: https://example.com/video.mp4

Step 2: Validate the Input

For Local Files:

1. Resolve relative paths to absolute paths
2. Verify the file exists
3. Check file extension is supported: .mp4, .mov, .avi, .mkv, .webm

For URLs:

1. Verify URL starts with http:// or https://

Step 3: Start Embeddings Task

Use the mcp__twelvelabs-mcp__start-video-embeddings-task tool:

For Local Files:

Tool: mcp__twelvelabs-mcp__start-video-embeddings-task
Parameters:
  videoFilePath: "<absolute-path-to-video>"

For URLs:

Tool: mcp__twelvelabs-mcp__start-video-embeddings-task
Parameters:
  videoUrl: "<url>"

Step 4: Report Result

On success:

Video embedding started!

Source: <filename-or-url>
Task ID: <task_id>

Embedding creation runs in the background and may take several minutes.
Ask "are my embeddings ready?" to check the progress.

Checking Embedding Status

When the user asks about embedding status ("are my embeddings ready?", "check embedding status"):

Step 1: Check Local Config

python3 -c "
import sys
sys.path.insert(0, '.twelvelabs')
from config_helper import get_all_pending_embedding_tasks
import json
print(json.dumps(get_all_pending_embedding_tasks(), indent=2))
"

Step 2: Query the API

If there are pending tasks, check each one:

Tool: mcp__twelvelabs-mcp__get-video-embeddings-tasks
Parameters:
  taskId: "<task-id>"

If no pending tasks, check the latest tasks:

Tool: mcp__twelvelabs-mcp__get-video-embeddings-tasks
Parameters: (none)

Step 3: Auto-Retrieve When Ready

If a task status is ready, automatically retrieve the embeddings:

Tool: mcp__twelvelabs-mcp__retrieve-video-embeddings
Parameters:
  taskId: "<task-id>"

Display the results to the user.

Retrieving Embeddings from an Indexed Video

When the user asks for embeddings of an already-indexed video ("get embeddings for video abc123"):

Step 1: Identify the Video

The user should provide a video ID. If not, help them find it:

• Check local config for known videos
• Use mcp__twelvelabs-mcp__list-videos to list available videos

Step 2: Retrieve Embeddings

Tool: mcp__twelvelabs-mcp__retrieve-video-embeddings
Parameters:
  videoId: "<video-id>"
  indexId: "<index-id>"  (optional, uses default index if omitted)

Step 3: Display Results

Show the embedding information:

Embeddings retrieved for video: <video-id>

- Number of embeddings: <count>
- Embedding dimension: <dimension>

[Display embedding details or summary]

Example Interactions

User: "Create embeddings for /Users/me/Videos/demo.mp4" Action: Validate file exists -> call start-video-embeddings-task with videoFilePath

User: "Embed this video https://example.com/video.mp4" Action: Call start-video-embeddings-task with videoUrl

User: "Are my embeddings ready?" Action: Check pending embedding tasks -> query API -> auto-retrieve if ready

User: "Get embeddings for video abc123" Action: Call retrieve-video-embeddings with videoId

Important Notes

• Async Processing: Embedding creation runs in the background on TwelveLabs servers
• Processing Time: Can take several minutes depending on video length
• Two Retrieval Paths: Embeddings can be retrieved from a task (using taskId) or from an indexed video (using videoId + indexId)
• Auto-Retrieve: Status checks automatically retrieve embeddings when tasks complete