The command string is still yours to write — nothing here guesses at filters — but the queue, the GPU workers and the hosting of the output are not your problem, and a finished job announces itself over a webhook instead of being waited on.
A remote MCP server in front of Very Good FFmpeg, a service that runs FFmpeg commands behind an API. You give it input URLs, output names and the command string; it queues the job and returns URLs for the files it made.
- Any FFmpeg command run as a queued job, with a per-job timeout you set
- Inputs read from public URLs; media that is not already hosted goes through the temporary files route first
- Several commands chained in sequence on one machine, for pipelines that should not round-trip between steps
- Jobs routed to CPU workers or to NVIDIA workers when the work wants a GPU
- A webhook when the job finishes, so nothing has to sit and poll
- The same account reachable from the TypeScript and Python SDKs, Make and n8n, if part of the pipeline lives outside the chat
A Very Good FFmpeg API key from the dashboard, or OAuth if your client supports the flow. It is a remote server over streamable HTTP: the key goes in an Authorization header, and clients that cannot set headers pass it as a query parameter instead — a form the vendor asks you to keep private, since URLs end up in client history, logs and screenshots.
One command plus a key — npx -y mcp-remote https://verygoodffmpeg.com/api/mcp, then supply credentials