Model Context Protocol
Ask your assistant to find a moment. BitterClip pulls up the transcript, opens the editor in the conversation, renders the MP4, and keeps the timing tied to the audio.
The assistant picks words. BitterClip handles the cut, captions, export, and final approval without guessed timestamps.
Start with one recordingLive · running below
The transcript editor below is the exact BitterClip editor that opens inside ChatGPT and Claude. It is mounted here so you can try the same review surface: drag across the words to cut a clip; press play and watch the cursor land on the word.
One editor. Wherever you are already working.
Fixtures only — every edit is stubbed, no backend, no account, no render cost. Point the same component at a live backend and it does the real work.
ChatGPT App Directory
The public app should be named BitterClip and submitted from a verified BitterClip or Bitter Platform organization. Its job is narrow: find source-linked clip candidates from a user's recordings, open the transcript editor, create approved clips, export MP4s, and prepare publishing packages only when the user asks.
POST /mcp) The BitterClip backend speaks JSON-RPC 2.0 over HTTP — CORS, origin checks, and bearer-token auth. Tool calls route to your workspace.
MCP over JSON-RPC 2.0, plus the MCP Apps UI extension (stable spec 2026-01-26) that mounts interactive widgets inside the chat.
initialize — Handshake and exchange capabilitiesping — Connection heartbeat verificationtools/list — Expose available editor capabilitiestools/call — run a tool — find moments, create or export a clipresources/list — Ingested audio/video resourcesresources/read — read a recording's transcript and clips{
"jsonrpc": "2.0",
"method": "tools/list",
"params": {},
"id": 1
}