samson-art/transcriptor-mcp

**Connect one server. Then ask Claude, ChatGPT or etc about a video:** the transcript, the chapters, the metadata, or a single frame. It works with 11 platforms, not only YouTube.
**Connect · What to ask · Widgets · Platforms · Self-host**
---
⚡ Connect in 30 seconds
The hosted endpoint is:
🖱️ One click
⌨️ One command, for Claude Code
Then run /mcp and approve the sign-in in the browser. After this, claude mcp list shows ✔ Connected.
🧭 No terminal
| Client | What to do |
| --- | --- |
| **Claude** (web and desktop) | Open Settings → Customize → Connectors. Select **Add** → **Add custom connector**, paste https://transcriptor.gateway.mcpal.io/mcp, then select **Add**. |
| **ChatGPT** | Open Transcriptor in the ChatGPT plugin directory and select **Install plugin**; sign in when asked. Or in ChatGPT open **Plugins**, search **Transcriptor**, select **Install plugin**. Then mention @Transcriptor in a chat. |
| **Codex** | Same directory, one install: ChatGPT and Codex share it. In a Codex task open **Sources** → **Use plugins** → **Transcriptor**; in the CLI, /plugins. |
> **Note:** a new directory listing can take up to 6 hours to appear in Codex (Plugins in ChatGPT and Codex).
🧩 Any other MCP client
If your client is not in the list above, add the server with this configuration:
If you want to run the server yourself, read Self-host. The tools are the same and you need no account.
---
🧰 What you can ask
| Ask for this | Tool |
| --- | --- |
| *"Summarize this video for me"* | get_transcript |
| *"Give me the subtitles as an SRT file"* | get_raw_subtitles |
| *"Is there a German track for this video?"* | get_available_subtitles |
| *"Who published this and how many views?"* | get_video_info |
| *"Go to the part about pricing"* | get_video_chapters |
| *"Show me the screen at 4:12"* | get_video_frame |
| *"Get English transcripts for the first 5 videos in this playlist"* | get_playlist_transcripts |
| *"Find recent videos about X"* | search_videos (YouTube) |
Long transcripts come in parts. Each response gives a cursor for the next part, so no text is lost.
Full tool reference (input and structured response)
Each tool that takes a video accepts url. This is a link from a supported platform or a plain YouTube ID. Each tool returns content (text for the chat) and structuredContent (typed JSON for your code).
get_transcript
Clean plain text, without timestamps, HTML, or speaker names. Without lang, the tool returns the track in the video's original language. Most platforms other than YouTube do not say which language a video is in; when the tool cannot tell which track that is, it answers with the list of tracks, and you call it again with type and lang. The inputs are the same as for get_raw_subtitles.
Response: videoId, url (the video page, as the server resolved it), type, lang, text, is_truncated, total_length, start_offset, end_offset. When more text is available, the response also has next_cursor.
get_raw_subtitles
Raw SRT or VTT content, in parts.
Input:
- type — official or auto. Without lang, the tool picks a track of this type
- lang — a language code or track name, as get_available_subtitles lists it. Without it, the video's original language, as for get_transcript
- response_limit — default 50000, minimum 1000, maximum 200000
- next_cursor — the cursor of the previous response
Response: the fields of get_transcript, plus format (srt or vtt) and content.
get_available_subtitles
Response: official and auto. Each field is a sorted list of language codes. Use this tool first, then give type and lang to the tools above.
get_video_info
Extended metadata from yt-dlp:
- identity — videoId, title, description, webpageUrl
- author — uploader, uploaderId, channel, channelId, channelUrl