Feeding Video to an LLM: Timestamped Transcripts over REST and MCP

Captions first, speech recognition as the fallback, and how to handle the async path in code