obsidian-mcp-brain is a thin Node.js HTTP server that gives remote MCP clients, such as Claude Code and Claude Desktop, access to one or more Obsidian vaults. It implements every tool natively and reads and writes the vault files directly.

Why it exists

Remote MCP clients connect over HTTP, not stdio, and the usual approaches have problems:

• No HTTP transport for local MCP servers. Wrapping a stdio MCP server with a proxy such as mcp-proxy can leave tools/list timing out: Claude Code opens its GET /mcp notification stream at the same moment it asks for the tool list, and the proxy’s response routing gets confused. This server implements the MCP HTTP transport directly and has no child process.
• No OAuth 2.0 discovery. The MCP spec requires every non-localhost remote server to expose OAuth 2.0 discovery endpoints, and Claude Code refuses to connect without them. This server serves a public, no-credentials OAuth flow so the handshake completes.

How it fits together

Claude Code / Claude Desktop
        │  HTTPS (e.g. Tailscale)
        ▼
obsidian-mcp-brain  :3002
  ├─ OAuth 2.0 discovery endpoints
  ├─ MCP Streamable HTTP transport (POST /mcp, GET /mcp)
  ├─ Path deny list (access control)
  └─ Vault access (node:fs/promises)
        │
        ▼
Obsidian vault (filesystem)

Where to go next

• Install the server and run it as a service
• Configure your vaults and access rules
• Connect a client such as Claude Code or Claude Desktop
• Browse the tools it exposes
• Read about the security model before exposing it on a network