Getting your content to an AI
An AI is only as personalized as the context it can reach. Most of anyone's best thinking happens in conversation, which is exactly the material that normally never gets written down. Carbon Voice captures it as a byproduct of communicating, which makes it unusually good raw material for an assistant that's meant to know you.
There are two ways to get it there, and they're genuinely different.
Directly, over MCP
Your assistant queries Carbon Voice at the moment you ask the question.
- Always current — it sees what you said five minutes ago.
- Nothing is stored anywhere new.
- Only works in MCP-capable clients — Claude, ChatGPT, Cursor, Windsurf.
Set it up: Connecting Claude, ChatGPT, Cursor, Windsurf.
Via a vault you own
The Obsidian sync plugin pulls your conversations and voice memos into an Obsidian vault as Markdown.
- A snapshot, refreshed on the interval you choose — not live.
- Anything that reads files can use it, not just MCP clients.
- It composes with the rest of your notes, so Carbon Voice content sits alongside everything else you've written and gets linked into it.
- The files are yours, and stay yours.
This is the "second brain" pattern: capture by voice in Carbon Voice, sync into the vault, then point an assistant at the vault so it answers with fuller context of you.
What lands in the vault matters here. It's transcripts, AI summaries, and participant and workspace links, not just audio. That's what makes it answerable later.
Which one
Use MCP when you want live answers about recent conversations inside an assistant you already talk to.
Use the vault when you're building something durable, a knowledge base you'll annotate and link, that outlives any one AI tool.
They aren't exclusive. Plenty of people run both.