·
1 commit
to main
since this release
Second Brain — Claude Desktop one-click add-on (v0.1.0)
Double-click installer for the Second Brain MCP server (@travisbergen2/second-brain-mcp@0.1.0) — your own notes, searched and cited by Claude Desktop, entirely on your machine.
Install: download second-brain-0.1.0.mcpb → double-click → Install → answer two questions (where your notes folder is; which folders inside it Claude may read). Change either anytime in Claude Desktop's Settings → Extensions.
Verification (2026-09-01)
server/index.jsinside the bundle is byte-identical to the published npm dist — sha256a50dc475c04c8cc387e5525a3010769be6a7a6c44d0877fc3cab0addfc85c163- Bundle sha256
5f1e4621f7f3bcbe229b7fd63ce22aa7405c1eeb2c4f534c35f708aaa025c094(mcpb tool shasum6daa728f7f2c38098790319a43dc51b25d67249d) - Manifest validated with
@anthropic-ai/mcpb(spec 0.3) - Smoke-tested 9/9 over a live MCP stdio handshake: allowlist mode (search → capped snippets with per-call disclosure; dated write-back note wikilinking its source; vault-side audit log), plugin-settings mode (Obsidian plugin allowlist honored live and honestly reported), and default-OFF mode (refusal with opt-in steps; zero note content shipped)
- Bundled dependencies:
@modelcontextprotocol/sdk@1.30.0,zod@3.25.76(Node ≥ 20; Claude Desktop's built-in runtime qualifies)
Privacy posture (unchanged from the npm package)
Folders OFF until listed · at most 6 snippets / 2,400 characters per search · every result discloses exactly which notes shipped · an audit note inside the vault keeps the receipts · writes are create-only and confined to the write-back folder · note content goes only to the AI app this is installed in. Nothing hosted, no telemetry.
Known limitations (v0.1.0)
- While this add-on is installed, its "Folders the AI may read" setting overrides the Obsidian plugin's live folder switch (
--allowprecedence). Plugin-follow mode is planned for v0.1.1. - Unsigned bundle: Claude Desktop may show an unsigned-extension notice at install.