Connect an AI assistant

MyStack runs a connector that lets an AI assistant you already pay for — Claude or ChatGPT — read your record and add new entries on your behalf, in your own words. You might say "log that I started 5g creatine this morning" or ask "what did my last ApoB read?" and have it land in your account. It runs on your AI subscription; MyStack never charges for it and never runs the model.

The connector endpoint is https://mystack.bio/api/mcp. Connecting takes you through a sign-in and a consent screen where you pick what the app may do.

Connecting an assistant means your data — including private entries — flows through your AI provider under their terms, outside MyStack's EU boundary. That's your call to make, and it is not the default.

If you'd rather keep your health data out of an LLM entirely, you don't need this at all: the paste importer and the editors in your dashboard do everything the connector does — add stack items, measurements, experiments, protocols — without any of it leaving for a third party.

A connected app can read everything in your account, private entries included, and add new entries — which always land private, exactly as if you'd typed them yourself. It can tell you what's public versus private, but it cannot change that.

It can never publish anything, delete anything, export your data, change your handle or email, or record consent on your behalf. Publishing health data is a deliberate human act that only happens in this interface, behind its own consent step — no assistant can do it for you.

In Claude (claude.ai), open Settings → Connectors → Add custom connector and paste the endpoint https://mystack.bio/api/mcp. Claude Desktop and Claude Code support the same custom connectors. Claude opens a browser to MyStack: sign in, then choose on the consent screen whether the app may only read, or also add entries.

Connecting is step one — in claude.ai chats there is also a per-chat toggle: in the chat input's tools menu, check that MyStack is enabled for that conversation (in Claude Code, /mcp shows the connection). And Claude decides when to reach for tools; if it already "remembers" something from your past chats it may answer from memory instead of your record. Saying "check my MyStack" makes it look. In the chat itself, a real lookup renders as an expandable tool block showing exactly what came back — a chat answer with no tool block came from memory, not from your record. One exception: artifacts (interactive apps Claude builds inside a chat) can use your connection without showing a tool block. Whatever the surface, anything an assistant adds lands private and marked as API-created, and every app you've granted stays listed — and revocable — under Settings → Connected apps.

ChatGPT speaks the same connector protocol, but availability is plan-gated — custom MCP connectors appear under developer mode and on business plans, and OpenAI has rolled them out unevenly. If you have access, add a custom connector pointing at https://mystack.bio/api/mcp and complete the same sign-in and consent screen. If you don't see the option, your plan doesn't expose it yet — nothing MyStack can change.

Every app you connect is listed under Settings → Connected apps, with the scopes you granted and when it was last active. Disconnecting there cuts off renewal immediately; any token already issued stops working within 15 minutes. Anything an app created stays in your account — disconnecting removes access, not your data.