InstaMCP
Self-initiated productA custom MCP server that gives Claude the video track: one public link returns a transcript and frame-by-frame visuals in a single call.
The problem
Claude could hear a video’s audio but not see the frames. A transcript alone loses the room, the light and the visual context that carries the line.
The build
I built InstaMCP, a custom MCP server that accepts a public Instagram link and returns a transcript plus frame-by-frame visuals in one API call. It runs on Cloudflare with auto-rotating keys.
The detail
The server uses faster-whisper for transcription and extracts frames from the video. The infrastructure is designed to cost zero, so the tool can stay free forever instead of becoming another paid API.
One call. Sound and picture.
Live at instamcp.dev-s.studio and free while it is early. One endpoint returns both tracks, with no second tool and no API key.