gitmyhub

brain.md

TypeScript ★ 25 updated 1mo ago

Local-first markdown notes with a first-class MCP server. 16 tools + 2 resources for Claude Code, Claude Desktop, Cursor and any MCP agent. Per-folder permissions, semantic RAG via LanceDB, single Bun binary. AGPL.

A local-first markdown note-taking app with a built-in server that lets AI agents like Claude read, search, and write your notes.

TypeScriptBunMCPsetup: easycomplexity 2/5

brain.md is a note-taking app and AI agent memory layer that runs entirely on your own computer. You write notes in standard markdown format, and the app gives you both a polished web editor and a built-in server that lets AI assistants read, search, and write those notes. Your files stay as plain text on disk, so you can open the same folder in other editors like Obsidian or VSCode at any time.

The app installs as a single binary with no Node.js or Python runtime required. Once running, it serves a web editor at a local port where you can write and preview notes with support for wikilinks, math rendering, task lists, code blocks, and diagrams. Every save is automatically committed to a local git history, so you can browse or restore earlier versions of any note. Optional password protection gates both the web interface and the AI connection.

The built-in MCP server (Model Context Protocol) is what makes brain.md useful for AI agents. It exposes 17 tools and 2 resources that agents like Claude Desktop, Claude Code, or Cursor can call directly, including searching your notes, reading a file, or writing a new one. You can set per-folder permissions so that sensitive directories stay out of reach for agents even while the rest of your notes remain accessible.

Semantic search is included out of the box. Notes are broken into chunks and converted to vectors using a small model that runs locally in your browser via WebAssembly, requiring no external database or internet connection. You can optionally point it at a cloud embedding service like OpenAI or Ollama instead.

The project is AGPL-licensed and positions itself as a simpler alternative to Obsidian or Logseq for people who primarily want a markdown vault designed from the start as a memory layer for AI agents, rather than a large plugin ecosystem.

Where it fits