菜单
Search indexed videos using natural language. Use when the user wants to find specific content, moments, or scenes in their indexed videos. Triggers on phrases like "find the part where...", "search for...", "look for...", "where does...", "when does...".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Upload, list, or delete TwelveLabs assets — generic reference files (images or videos) used by entities and async-analyze. Triggers on "upload this image", "upload this video as an asset", "delete that asset", or any explicit asset-management request.
Asynchronously analyze a video — by URL, local file path (auto-uploaded), asset ID, or base64 — no prior indexing required. Use for URL inputs, videos over 1 hour, time-based metadata extraction, multimodal prompting with reference images, structured JSON output, or when the user asks to check/list/delete async analysis tasks.
Set up and search for specific people or objects in videos using entity search. Use when the user wants to find a specific person, track someone across videos, or set up face/object recognition. Triggers on phrases like "find this person", "set up entity search", "track this person in my videos", "who is in this video".
Synchronously analyze a video to understand its content. Accepts an indexed videoId, a direct video URL, a previously uploaded asset, or a local file path (auto-uploaded). Use when the user asks "what is this video about?", "summarize the video", "analyze this video", or has questions about video content.
Create, check, and retrieve video embeddings. Use when user wants to generate embeddings from a video, check embedding status, or retrieve embeddings for a video.
Search videos using an image combined with text for refined results. Use when the user wants to search with a reference image, says "find videos matching this image", "search with this picture", or wants to combine visual reference with text description.
基于 SOC 职业分类
| name | search |
| description | Search indexed videos using natural language. Use when the user wants to find specific content, moments, or scenes in their indexed videos. Triggers on phrases like "find the part where...", "search for...", "look for...", "where does...", "when does...". |
Search indexed videos using natural language descriptions. TwelveLabs interprets your query to find matching content based on visual elements, actions, sounds, and on-screen text.
Use this skill when the user:
Identify what the user wants to find from their natural language request. Convert their question into a search query.
Examples:
Use the mcp__twelvelabs-mcp__search tool:
Tool: mcp__twelvelabs-mcp__search
Parameters:
query: "<extracted search query>"
Note: The search is performed across all videos in the default index. To search a specific index, provide the optional indexId parameter.
Format the search results clearly for the user. The search returns matching segments with timestamps.
For each video with matching segments, display:
Example output with results:
Found matches for: "a person walking"
**video_filename.mp4**
Stream: https://stream.url/video.m3u8
Matching segments:
- 00:12 - 00:28 (16 seconds)
- 01:45 - 02:03 (18 seconds)
- 03:30 - 03:45 (15 seconds)
**another_video.mp4**
Stream: https://stream.url/another.m3u8
Matching segments:
- 00:05 - 00:15 (10 seconds)
Found 4 matching segments across 2 videos.
Example output with no results:
No matching content found for: "a person walking"
Tips:
- Try different keywords or descriptions
- Use broader search terms
- Make sure your videos are fully indexed
Convert timestamps to human-readable format:
MM:SS (e.g., 01:45)HH:MM:SS (e.g., 1:23:45)Calculate segment duration as end_time - start_time and display in parentheses.
TwelveLabs uses multimodal AI to understand video content:
The search interprets natural language queries and finds segments that match across all these modalities.