sync-analyze

Synchronous Video Analysis

Analyse a video using TwelveLabs AI and return the result inline. Works on:

• A previously-indexed videoId (legacy, Pegasus 1.2 only).
• A direct http(s) URL (Pegasus 1.5).
• A previously-uploaded assetId (Pegasus 1.5).
• A local file path — this skill uploads it as an asset internally first, then analyses it (Pegasus 1.5).

For videos longer than 1 hour, or for time-based metadata extraction, use the async-analyze skill instead.

When to Use This Skill

The user wants a quick text/JSON answer about a single video and:

• Provides a video reference (URL, file path, asset ID, or indexed videoId) AND a question / instruction; or
• Asks a general question and references "this video" / "that video" / "the video I just indexed".

For multi-segment structured extraction (e.g. "find every product mention with timestamps"), use the async-analyze skill — that supports analysisMode: "time_based_metadata".

Model Default

Input type	modelName
videoId	pegasus1.2 (forced — the MCP rejects 1.5 + videoId)
Anything else	pegasus1.5

Only override to pegasus1.2 when the user explicitly says so ("use 1.2", "legacy model").

Instructions

Step 1: Classify the Input

Look at what the user gave you:

• 24-char hex string referred to as a "video" → videoId. Force modelName: "pegasus1.2".
• http(s)://... URL → videoUrl. Use modelName: "pegasus1.5".
• 24-char hex or asset_* referred to as an "asset" or "uploaded thing" → assetId. modelName: "pegasus1.5".
• Anything that looks like a file path → resolve to absolute; verify the file exists; upload it first:

  Tool: mcp__twelvelabs-mcp__create-asset
  Parameters:
    file: "<absolute-path>"
  

  Capture the returned assetId. Tell the user the asset was created and that they can reuse or delete it via /twelvelabs:assets. Then continue as if they had given you an assetId.

Step 2: Build the Prompt

If the user gave a question, use it as prompt. Otherwise default to: "Provide a comprehensive summary of this video including key moments, topics covered, and important details."

Step 3: Call the Analyze Tool

Set exactly one source (videoId, videoUrl, assetId, or base64Video) plus modelName and prompt:

Tool: mcp__twelvelabs-mcp__sync-analyse-video
Parameters:
  <one of: videoId | videoUrl | assetId | base64Video>: "..."
  modelName: "pegasus1.5"        # or "pegasus1.2" only for videoId / explicit user override
  prompt: "..."

Step 4: Advanced Parameters

Add these only when the user asks for the corresponding behavior:

• temperature (0-1, default 0.2) — "more creative" / "more deterministic"
• maxTokens — long outputs
• jsonSchema — "give me JSON"; provide a JSON Schema (Draft 2020-12); cannot combine with promptV2
• startTime / endTime (Pegasus 1.5 only) — clip to a sub-range; end - start ≥ 4
• promptV2 (Pegasus 1.5 only, mutually exclusive with prompt) — structured prompt with image references via <@name> and a mediaSources array (each: { name, mediaType: "image", url | assetId | base64String })

Step 5: Display

Video Analysis

<analysis output>

Video: <best identifier — filename, URL, asset ID, or videoId>
Model: pegasus1.5    (or pegasus1.2)

Step 6: Errors

• videoId is not accepted with modelName='pegasus1.5' → you misclassified; either drop modelName (server will fall back) or switch to pegasus1.2.
• Provide exactly one of videoId, videoUrl, assetId, or base64Video → you set zero or multiple sources; re-classify.
• Asset still processing → wait a couple seconds and retry (create-asset returns status pending initially).

Example Interactions

"What is this video about?" (after listing videos and the user picked one) → sync-analyse-video with videoId + Pegasus 1.2, generic summary prompt.

"Analyse https://example.com/clip.mp4 and tell me what happens" → sync-analyse-video with videoUrl + Pegasus 1.5.

"Summarise /Users/me/keynote.mp4" → create-asset with file: "/Users/me/keynote.mp4" → sync-analyse-video with the new assetId + Pegasus 1.5.

"What products are mentioned in this video? Return JSON: { product, timestamp_sec }[]" → sync-analyse-video with the appropriate source + jsonSchema set to an array schema.

Important Notes

• Indexed-video analysis (legacy) still works via videoId but is capped at Pegasus 1.2 capabilities.
• Asset-based analysis runs on Pegasus 1.5 — wider feature set (clip windows, structured prompts, big token budgets).
• Auto-uploaded assets persist. Always tell the user the asset ID so they can reuse or delete it.