Video Understanding MCP runs on your own machine — the client starts it, so there is no endpoint to ping. Last commit 9 Nov 2026.
Let AI agents watch videos and answer questions about them through any OpenAI-compatible endpoint.
Nothing serious here today
Today is the operative word: we check Video Understanding MCP every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
Connect this server
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add video-understanding-mcp -e 'VIDEO_BASE_URL=<VIDEO_BASE_URL>' -e 'VIDEO_MODEL=<VIDEO_MODEL>' -- npx -y @granitebps/video-understanding-mcp{
"mcpServers": {
"video-understanding-mcp": {
"command": "npx",
"args": [
"-y",
"@granitebps/video-understanding-mcp"
],
"env": {
"VIDEO_BASE_URL": "<VIDEO_BASE_URL>",
"VIDEO_MODEL": "<VIDEO_MODEL>"
}
}
}
}[mcp_servers.video-understanding-mcp]
command = "npx"
args = ["-y", "@granitebps/video-understanding-mcp"]
env = { VIDEO_BASE_URL = "<VIDEO_BASE_URL>", VIDEO_MODEL = "<VIDEO_MODEL>" }{
"mcpServers": {
"video-understanding-mcp": {
"command": "npx",
"args": [
"-y",
"@granitebps/video-understanding-mcp"
],
"env": {
"VIDEO_BASE_URL": "<VIDEO_BASE_URL>",
"VIDEO_MODEL": "<VIDEO_MODEL>"
}
}
}
}{
"servers": {
"video-understanding-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@granitebps/video-understanding-mcp"
],
"env": {
"VIDEO_BASE_URL": "<VIDEO_BASE_URL>",
"VIDEO_MODEL": "<VIDEO_MODEL>"
}
}
}
}Put your own values in place of
<VIDEO_BASE_URL> (Base URL of an OpenAI-compatible chat completions endpoint whose model accepts video, e.g. https://host/v1), <VIDEO_MODEL> (Model (or router alias) to send videos to).
The author’s registry entry marks them as required.
Optional settings from the registry entry, left out of the block above:
VIDEO_API_KEY (Endpoint API key, sent as a bearer token), VIDEO_PART_FORMAT (How the video is attached: video_url, image_url, file, or input_file), VIDEO_MAX_DURATION_SECONDS (Longest analyzed range in seconds), VIDEO_MAX_UPLOAD_MB (Largest prepared video sent to the endpoint, in MB), VIDEO_ALLOWED_DIRS (Directories local files may be read from; local files are refused when unset), VIDEO_VERIFY_MEDIA (Fail calls when the endpoint evidently ignored the video).
Alternatives to Video Understanding MCP
same job, measured the same wayAsk questions about company data, set up watchers on ongoing goals, and catch up on findings.
MCP server for image/video understanding & generation (Gemini/OpenAI/Grok)
Generate and edit images and video via OpenRouter or any OpenAI-compatible API.
MCP server for BEHAVR: remember images/videos and ask questions about them
Dispatch litigation work to legal vendors from any MCP-compatible AI workflow. Patent pending.
Connect Claude to any OpenAI-compatible LLM endpoint and offload routine work to a local model.
Evidence-backed codebase understanding with file:line citations, for AI clients over MCP.
Use Chinese LLMs through the SinoGate OpenAI-compatible API.
Video Understanding MCP — questions
Answers built from our own checks of this server.