Beyond CLAUDE.md
CLAUDE.md is great for project instructions Claude Code loads every session. Noesis handles the other half: the growing body of knowledge — research, decisions, snippets, things you learned — that's too big to paste into a config file. Those stay as ordinary markdown notes, and the MCP server lets Claude Code pull exactly the right one on demand instead of you copy-pasting it in. Both speak plain markdown; nothing is locked in.
Plain files vs Noesis
strong· partial / depends· not built in
Where plain files still win
Honest concessions: if your notes already live in the repo Claude Code is working in, plain files need zero setup and no server. Noesis earns its keep once you have more notes than you can grep, want them on your phone, or want semantic search and sync. Keep both — Noesis reads and writes the same files.
Built around Claude Code via MCP
Noesis ships a first-party @noesis-brain/mcp-server on npm so Claude Code can read, write, search, and sync your markdown notes directly from the terminal. 19 tools cover semantic search, embeddings, conflict-aware 3-way merge for offline sync, AI metadata enhancement, and related-note discovery. Plug it into Claude Code in two lines of config and your notes become a first-class context source for every prompt.
$ npm install -g @noesis-brain/mcp-server
# Then from Claude Code
$ claude
# > What did I learn about auth?
search_semantic("auth flow")
No lock-in — it's just markdown
Your .md files import as-is and export as-is. Point Noesis at a folder, use it from Claude Code, and if it doesn't click your files are exactly where you left them — plain markdown, no proprietary format.
