Expose Complete YouTube Transcripts Through MCP
E
Eric Poulin
Hi Recall team,
I’ve been testing the Recall MCP integration with YouTube cards and found a useful enhancement opportunity.
Recall already extracts the full YouTube transcript and lets me export it from the Recall interface. Through MCP, however, an AI client can currently access only selected transcript chunks through search or get_document_content.
The MCP does expose timestamped reader chunks, alongside the generated notebook summary chunks, which is great. But get_document_content returns a limited, non-contiguous selection rather than the complete transcript. I can retrieve additional sections through targeted searches, but I cannot deterministically reconstruct or export the entire transcript.
Would you consider adding a dedicated transcript operation, perhaps something like:
get_transcript(card_id, format, include_timestamps, cursor/page_size)
Ideally, it would:
* Return the complete source transcript, separate from Recall’s summaries
* Preserve timestamps
* Support pagination for long videos
* Maintain deterministic transcript order
* Indicate whether the transcript is complete or partially available
* Optionally return plain text, Markdown, WebVTT, or SRT
This would make Recall particularly valuable as a source repository for AI research workflows. An assistant could retrieve the original transcript for detailed analysis, quotations, fact verification, or creating structured notes, while still using Recall’s summaries for faster discovery.
I tested this with the YouTube card today. The MCP exposed transcript excerpts from approximately 00:00 through 14:36, but only as selected chunks rather than one complete, sequential transcript.
Happy to share more details from the test if useful.
Log In
Sankari Nair
Hi Eric, this is Sanks, one of the co-founders. This is a fantastic feature request and makes perfect sense. I will raise it with the team, and hopefully we can get back to it quickly.
E
Eric Poulin
Sankari Nair, thank you so much for your answer. I am planning to create a small agent in Codex to look that a YouTube video I share into Recall, get the link via source_url, and use an external tool to generate the transcript. The funniest thing is that I found that Recall have a Free YouTube Transcript Generator that I think will work very well! :) ("https://www.recall.it/youtube-transcript-generator") .
I would rather have it in Recall via MCP however :) Thanks for looking into it!
Sankari Nair
Eric Poulin: Eric, I spoke to the team on this one, and we are going to do a full revamp of our MCP, hopefully by the end of this month/early next month. It will also include write capabilities, and we'll be sure to include this as well. I am sorry that it'll take a couple of weeks for us to get there, but I think you'll have a far superior experience when we are ready.
E
Eric Poulin
Sankari Nair, thank you for the update! This is great to hear and looking forward to it. I would be happy to beta test if needed at some point!
If they are planning to do a full revamp, allow me to "abuse" a bit with another request. Would it be possible to add the possibility to manipulate tags via MCP? Create, Assign, Remove? This would fit very well with my agentic workflow to manage my second brain, as it would allow to classify my entries in recall but also across multiple system.
Thanks again for taking the time to look at my request, very cool!