MCP, and why your personal knowledge graph is suddenly worth something
Model Context Protocol turns your private corpus into a first-class citizen for Claude, Cursor and every agent shipping next year. Here's what that means for the way you save.
By BrainTube Team · · 8 min read
For two years 'my notes' had a single consumer: me. The Model Context Protocol changes that. Claude, Cursor, ChatGPT desktop and a growing pile of agents can now read your library directly — which means the way you save in 2026 has to assume a non-human reader.
What MCP actually does
MCP is a thin standard for exposing a tool or a data source to a language model. The model asks 'do you have anything about X?', your server replies with the matching chunks plus citations, and the model writes back an answer that's grounded in your corpus instead of its training data. For knowledge tools, this is the equivalent of getting an API the day before everyone needs one.
Why your old notes are bad context
A 2,000-word Notion page with no headings, no quotes, no source links is almost useless as context. The model gets a wall of prose and either ignores it or hallucinates around it. The same content as 12 chunked claims with timestamps and source URLs is gold — the model can quote it, cite it, and stay grounded. Saving for LLMs means saving in chunks.
What we changed in BrainTube
Every Brain in BrainTube is exposed as an MCP server you can paste into Claude or Cursor in 30 seconds. Each chunk carries its source video, channel, timestamp and your highlight colour. That's enough metadata for the model to write 'According to <channel>, at <timestamp>, the argument is X' — which is what you want a research assistant to do.
The 2026 default
By the end of this year, every serious knowledge tool will speak MCP or something like it. The interesting question is no longer 'where do I save?' but 'what shape am I saving in?' Atomic, sourced, chunked. Anything else and your agents will quietly route around you.
Takeaways
- MCP makes your private corpus first-class context for any agent.
- Unstructured prose is bad context — chunked, sourced claims are good context.
- Every BrainTube Brain is already an MCP server.
Try BrainTube on your own corpus
Free tier, no card. Export anytime.
More to read
- Second Brain vs AI Memory Layer: What Actually Changed — A second brain system organises what you write. An AI memory layer makes what you consume queryable by any model. The difference decides which one you should actually build.
- Stop watching videos. Start querying them. — The 60-minute video you bookmarked won't get watched. But the four answers buried in it will get used — if you can pull them on demand.
- How to build a second brain from your YouTube history — A practical workflow for turning years of passive watching into a knowledge graph you can actually use.