Can I Control What Claude Sees When I Connect HealthOS

Yes. You control this on the consent screen before the connection is approved. Reading is ticked by default, so Claude can see core health data such as activity, sleep, glucose, meals and blood tests after you approve. Logging data for you, DNA, Ask, community and account changes stay off until you tick them. Deletion is not simply off; anything that deletes asks for confirmation first.

Can I control what Claude sees when I connect HealthOS?

Yes. The permission list appears before the connection is approved, not after. You tick what to share, and the grant creates a token with only those scopes. The consent screen says reading is on by default and everything else starts off. What you leave unticked is not in the token Claude uses. Claude only gets what you ticked, and only when a question needs it.

You control this at connection time: read categories are ticked by default, everything else stays off, and anything that deletes data asks first.

What is on by default and what stays off

The default is not one share-everything switch. The read family is on. That includes activity, sleep, glucose, meals, blood tests, and the other read categories shown on the consent screen. Everything that changes data is off.

• Reading: on by default. Claude can answer questions about data the iPhone app has synced.
• Logging data for you: off. This is the write switch. Without it, Claude cannot log a weight, meal, water or goal.
• DNA: off. If you tick it, Claude gets findings and lookups, never the whole genome.
• Ask, community, export and account changes: off until ticked.
• Deleting: destructive tools also ask you to confirm before they run.

That list matters because a token is not a vague "Claude can see my stuff" grant. Each scope is a separate permission, and every tool call checks the scope on the token.

What "shared by default" looks like in real data

With default read access, Claude can answer "Summarise my last blood test and what changed since the one before" or "Plot my glucose after dinner for the last 7 days." It reads numbers, dates, reference ranges and the timestamp of the last phone sync. It cannot create an entry, change a goal, share your data or delete a reading unless you have ticked the matching scope and, for destructive actions, confirmed.

If you ask Claude to log 500 ml of water without write access, it cannot do it. If you ask it to delete a meal, it should refuse unless you both have write access and confirm. This is the difference between read and write, and it is applied per tool call, not per chat.

What stays limited even after you tick nearly everything

Some limits are built in. DNA tools return findings and lookups, never a raw genome file. Payment tools do not exist over OAuth, so an assistant cannot buy anything. Sharing tools are separate and ask before running. Even a broadly permissioned connection does not become an account admin.

Apple Health and glucose data come from the iPhone app. Claude sees what the phone last synced, and each record carries a sync time. If the phone has not synced since yesterday, a graph ends yesterday. Ask Claude when the data was last synced if you need to know whether you are looking at yesterday or just now.

How long access lasts, and how to stop it

A connection lasts at most 365 days, then the app asks again. Access tokens last one hour and refresh tokens rotate on every use, so after you disconnect, access stops quickly. You can disconnect any time in Settings → Connected apps, in the app or on the web. The limits are 60 tool calls a minute and 2,000 tool calls a day per connection. That is enough for normal chat queries and not enough for a background scraper.

To see this in your own numbers, use the HealthOS connector. It has the step-by-step Claude setup and what you can ask.