Video Understanding MCP

Local install

io.github.granitebps/video-understanding-mcpv1.0.1 GitHub verified

Let AI agents watch videos and answer questions about them through any OpenAI-compatible endpoint.

Design & Creative

Watch Source

Registry entry

Open in the registry

{
  "_meta": {
    "io.modelcontextprotocol.registry/official": {
      "status": "active",
      "isLatest": true,
      "updatedAt": "2026-11-09T10:41:27.620639Z",
      "publishedAt": "2026-11-09T10:41:27.620639Z",
      "statusChangedAt": "2026-11-09T10:41:27.620639Z"
    }
  },
  "server": {
    "name": "io.github.granitebps/video-understanding-mcp",
    "_meta": {
      "io.modelcontextprotocol.registry/publisher-provided": {
        "buildInfo": {
          "commit": "d9d10ad5c5600fb5510e4e95c0a89c3443495955"
        }
      }
    },
    "title": "Video Understanding MCP",
    "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
    "version": "1.0.1",
    "packages": [
      {
        "version": "1.0.1",
        "transport": {
          "type": "stdio"
        },
        "identifier": "@granitebps/video-understanding-mcp",
        "registryType": "npm",
        "environmentVariables": [
          {
            "name": "VIDEO_BASE_URL",
            "format": "string",
            "isRequired": true,
            "description": "Base URL of an OpenAI-compatible chat completions endpoint whose model accepts video, e.g. https://host/v1"
          },
          {
            "name": "VIDEO_MODEL",
            "format": "string",
            "isRequired": true,
            "description": "Model (or router alias) to send videos to"
          },
          {
            "name": "VIDEO_API_KEY",
            "format": "string",
            "isSecret": true,
            "description": "Endpoint API key, sent as a bearer token"
          },
          {
            "name": "VIDEO_PART_FORMAT",
            "format": "string",
            "default": "image_url",
            "description": "How the video is attached: video_url, image_url, file, or input_file"
          },
          {
            "name": "VIDEO_MAX_DURATION_SECONDS",
            "format": "number",
            "default": "300",
            "description": "Longest analyzed range in seconds"
          },
          {
            "name": "VIDEO_MAX_UPLOAD_MB",
            "format": "number",
            "default": "15",
            "description": "Largest prepared video sent to the endpoint, in MB"
          },
          {
            "name": "VIDEO_ALLOWED_DIRS",
            "format": "string",
            "description": "Directories local files may be read from; local files are refused when unset"
          },
          {
            "name": "VIDEO_VERIFY_MEDIA",
            "format": "boolean",
            "default": "true",
            "description": "Fail calls when the endpoint evidently ignored the video"
          }
        ]
      }
    ],
    "repository": {
      "url": "https://github.com/granitebps/video-understanding-mcp",
      "source": "github"
    },
    "description": "Let AI agents watch videos and answer questions about them through any OpenAI-compatible endpoint."
  }
}

Maintain this server? Claim it

Try it

This server only runs locally, from a package. Install it with the instructions on this page to try it.

Install

Pick your client. Values in capitals, like YOUR_TOKEN, are yours to fill in.

One click: Add to Cursor Add to VS Code Add to LM StudioThen replace the YOUR_ values and {placeholders} with your own.

Run in a terminal

claude mcp add video-understanding-mcp \
  -e VIDEO_BASE_URL=YOUR_VIDEO_BASE_URL \
  -e VIDEO_MODEL=YOUR_VIDEO_MODEL \
  -- npx -y @granitebps/[email protected]

Environment

VIDEO_BASE_URL (required): Base URL of an OpenAI-compatible chat completions endpoint whose model accepts video, e.g. https://host/v1
VIDEO_MODEL (required): Model (or router alias) to send videos to
VIDEO_API_KEY (optional, not in the config above): Endpoint API key, sent as a bearer token
VIDEO_PART_FORMAT (optional, not in the config above): How the video is attached: video_url, image_url, file, or input_file
VIDEO_MAX_DURATION_SECONDS (optional, not in the config above): Longest analyzed range in seconds
VIDEO_MAX_UPLOAD_MB (optional, not in the config above): Largest prepared video sent to the endpoint, in MB
VIDEO_ALLOWED_DIRS (optional, not in the config above): Directories local files may be read from; local files are refused when unset
VIDEO_VERIFY_MEDIA (optional, not in the config above): Fail calls when the endpoint evidently ignored the video

Tools

We don't run packages, so we can't list this server's tools.