summarize_video

Generate a structured video summary that covers key scenes, topics discussed, and a rough timeline. Useful for meetings, lectures, product demos, and any long-form content.

Parameters

Parameter	Type	Required	Description
videoPath	string	✓	Path to the video, file:// URI, or remote URL
provider	string	—	Override the AI provider: gemini, m3, kimi, qwen, mimo, glm

Usage Examples

Summarize a local video:

“Summarize ./long-video.mp4”

Summarize a meeting recording:

“Give me a structured summary of this meeting: ./meeting-2024-01-15.mp4”

Summarize a YouTube video:

“Summarize this YouTube video: https://www.youtube.com/watch?v=dQw4w9WgXcQ”

Use a specific provider:

“Summarize ./lecture.mp4 using Kimi”

Keyframe Extraction

The VIDEO_MCP_MAX_FRAMES environment variable controls the frame-based summarization path:

• Gemini uploads the full video natively, so VIDEO_MCP_MAX_FRAMES does not apply.
• MiniMax-M3 is image-only, so provider: "m3" always uses extracted keyframes.
• GLM, Qwen, and MiMo use direct URLs or base64 when possible and fall back to keyframes when provider size limits require it.
• Kimi prefers native upload up to 100 MB and only falls back to keyframes for large long videos that exceed its upload limit.

When the frame path is used, these defaults apply:

Tier	Default	Maximum
Free	50 frames	50 (clamped)
Pro	100 frames	Uncapped (set 0 for unlimited)

Frames are automatically trimmed to fit the active provider’s byte and image-count limits. Gemini uploads the full video natively, so keyframe limits do not apply there.

S3 relay bypasses these limits

When S3 relay is configured (AWS_S3_BUCKET set) and the provider is GLM, Qwen, or MiMo, the video is uploaded to S3 and a presigned URL is passed directly to the provider — regardless of video length or file size. No keyframe extraction runs, so VIDEO_MCP_MAX_FRAMES does not apply.

The frame limits still apply when using MiniMax-M3 (which is always frame-based), when S3 relay is not configured, for Kimi videos larger than 100 MB (which fall back to frames), or if the S3 upload fails and the server falls through to the frame-extraction path.