Second Brain for AI

Second Brain speaks the Model Context Protocol, so it works as a shared MCP memory server for Claude, ChatGPT, Cursor, Codex, and other compatible clients. They all read and write the same store through a single address. Nothing to copy or sync between apps. The trade-off is that a client has to support MCP or one of the capture surfaces. A tool that talks to none of them stays outside the loop, and for that tool, the memory simply doesn't exist yet.

Search runs on stored meanings rather than keyword matches. Time-aware ranking pushes the most current version of a fact to the top. Ask for what you meant and the note turns up even if you wrote it with different words months ago, which is the whole reason semantic recall beats a plain search box for personal notes. Old versions stop crowding out the answer you need now. That saves real time over scrolling.

When you save something that overlaps an existing memory, the system compares the two by meaning before storing. Very similar captures are blocked as duplicates, and near-duplicates go to a merge decision. Newer updates win the answer while the older record stays in the history. A moved meeting time won't leave your notes arguing with each other. Small detail. Big relief.

Every memory can be browsed, edited, appended to, connected, shared, exported, or permanently removed from the web dashboard. Each entry shows where it came from, and high-importance notes are protected from being quietly overwritten. Export comes out as JSON and Markdown with restore. That matters if you ever want to leave. No lock-in here.

Team Edition runs on the same Worker at no extra charge. Each person gets a personal workspace nobody else can read, alongside a shared layer that only fills when someone deliberately shares. Sharing moves one canonical memory rather than making a copy. Only the author or an admin can edit, delete, or un-share it. Admins manage people and integrations without gaining access to anyone's private memories, which is the part most team tools get wrong.

Memories can be saved from MCP clients, a CLI, a browser extension, an Obsidian plugin, Notion, calendars, email, iOS Shortcuts, or the dashboard itself. Calendar and inbox sync run read-only, so the week's movements land by themselves without you filing anything. Why does that matter? Because a memory layer only earns its keep if saving costs you nothing in the moment. This spread of capture surfaces is what makes it useful rather than another inbox to maintain. Save from the terminal. Save from the share sheet. It all lands in one store.

You can file memories into named projects, like a website, a client, or a room. A question about one returns only that project's notes. The tools sort captures into projects for you. Optional weekly insights gather themes recurring across what you've stored, which saves you compiling them by hand. That last part is genuinely useful once the store gets big.