MCP

MCP is Anthropic's open standard for letting AI models read from and write to external systems through a uniform interface. Model Context Protocol. It launched quietly in late 2024 and has since become the default way Claude (and increasingly ChatGPT, via OpenAI's compatible work) connects to real data. Your filesystem, your tools, your knowledge base, your APIs.

The posts here cover what MCP actually does, why it matters for anyone running a personal knowledge base, and how Minibase Vault uses it to make every saved article instantly available as context to Claude or ChatGPT. Without MCP, "AI with access to your notes" meant uploading files manually, hoping the model didn't hallucinate, and re-uploading next session. With MCP, the model has a stable read connection to your real vault. No uploads, no copy-paste, no per-session re-priming.

Some posts are conceptual: what MCP is, how it differs from the embedding and RAG patterns that came before. Others are practical: how to set up Minibase Vault as an MCP server, what it looks like in Claude Desktop, the trade-offs between giving an agent full vault access versus scoped access. If you've been running an Obsidian or Logseq vault for a while and the question is "how do I plug Claude into this without rebuilding everything", MCP is the answer and these posts are the walkthrough.