100 Requests a Day vs Unlimited Local Reads
A hosted health connector hands you a daily request allowance. A local Apple Health MCP server hands you the file. Once you have done the arithmetic on how many tool calls a week of real health questions actually takes, the allowance stops looking generous.
Quick answer
A hosted connector meters you by request: a documented daily cap on how many times your agent may call the service. A local Apple Health MCP server reads a JSON file on your own machine, so there is no request to count, no cap to hit, and no vendor between you and the file. If you ask your agent twenty questions in an evening, a metered connector spends twenty (or more, since one question can fan out into several calls). A local server spends nothing but battery.
What a request quota is really capping
The meter is not on your data. It is on the round trip. Every time your agent wants a number, a hosted connector has to call a server that has to read a copy of your health data it is holding. That round trip is what gets counted. It is a clean way to run a service, and it is also the reason a health conversation gets rationed.
Compare that to the file path. When your agent asks a local server for your resting heart rate over the last 30 days, the server opens a JSON file on the same machine, filters it, and answers. There is no network hop, so there is no unit for a vendor to bill.
The arithmetic of a week of health questions
People underestimate how fast tool calls add up. Here is a worked count for a single week of ordinary use, no exotic set up:
- Morning check: sleep duration, sleep stages, resting heart rate. Three questions, but a decent agent calls
get_metriconce per metric, so three calls. - Post workout: heart rate recovery and HRV against baseline. Two more calls, plus a
compare_periodscall. Three calls. - Evening: steps and active energy trend for the day. Two calls.
- Once a week: a trend sweep across sleep, HRV and resting heart rate.
get_trendsper metric, three calls, plus onecorrelate_metricscall. Four calls.
That is 12 calls on a normal day and 16 on the trend day. Seven days of that is roughly 88 calls a week, about 380 a month. Double it if you and a partner both ask, or if you reframe a question and the agent retries. You are now at or past a 100-a-day cap on a heavy day, and every experiment you want to run eats the allowance you wanted for tomorrow.
On a local server the same week costs 380 file reads and zero requests. You can ask the same question five different ways to see which phrasing gets the best answer, and nothing runs out.
What the metered tier actually costs
For the concrete shape of a request-metered health connector, here is the published pricing from the current market entrant we read on 2026-10-18: HealthAPI prices by request quota, with a Pro tier at $4.99 a month for 100 requests a day and a Developer tier at $9.99 a month for 1,000 a day, after a 2-day preview and a 7-day trial (evidence E1 and E2 in our weekly market pack). The point is not the exact figure. The point is that the number that decides whether you can use your own data is a quota, not your data.
How we sourced this. The quota tiers above come from the vendor's own connect page, read and hashed on 2026-10-18 (E1, E2). Quotas and prices change without notice, so check the source before you plan around a cap.
Where a quota changes your behaviour
The hidden cost of a cap is not the dollar figure. It is that you start self-censoring. You stop asking the exploratory question, because you might want the allowance later. You batch questions into one mega-prompt to save calls, which produces worse answers. You stop letting an agent pull a week of context, because a week of context is a week of calls.
Unlimited local reads change the shape of the conversation back to normal. Ask the follow-up. Ask for the raw numbers. Ask it to check a different metric because the first one looked interesting. The wall is gone.
The trade you are actually making
| Metered hosted connector | Local file MCP server | |
|---|---|---|
| Unit billed | Requests per day | None |
| Where data lives | Vendor-held copy | Your own disk |
| Offline | No | Yes, the file is already there |
| Best question you can ask | The cheapest one | The one you actually want answered |
| Cost shape | Recurring, per period | One purchase |
None of this makes a hosted connector a bad product. If you want zero set up and you do not mind a vendor holding a copy, it is a reasonable choice. It is a bad choice for the person who wants their own data, read as often as they like, on their own machine.
How the local path works
MetricBridge ships a zero-dependency stdio MCP server that reads your local Apple Health JSON export and exposes 16 read-only tools. Point a desktop agent at it with a small config block, and every read is a local file read.
{
"mcpServers": {
"health-export": {
"command": "node",
"args": ["/Users/you/.health-export-mcp/server.mjs"],
"env": {
"HEALTH_DATA_DIR": "/Users/you/Library/Mobile Documents/iCloud~ai~healthexport~app/Documents"
}
}
}
}Quit and relaunch the agent, ask it to call get_mcp_status, and it will tell you whether the server is up and whether it found your data. From there, ask away. Nothing counts the questions.
FAQ
Does a local Apple Health MCP server have a request limit?
No. A local server reads a JSON file on your own machine, so there is no network round trip and nothing to meter. You can call the tools as often as your agent will let you. What you are bounded by is your machine, not a vendor's quota.
How many tool calls does a normal week of health questions take?
In the worked count above, a typical day is about 12 calls and a trend day about 16, so roughly 88 calls a week and about 380 a month before retries. That is the number that makes a daily cap of 100 requests feel tight on a heavy day.
Is a request quota the same as a data limit?
No. The quota caps how many times your agent may contact the service, not how much of your data exists. Your Apple Health data is complete either way. The cap only decides how often you are allowed to look at a vendor's copy of it.
Can I switch from a metered connector to a local server later?
Yes, and it is a config change, not a migration. Export your Apple Health data once, keep the JSON on your machine, and point your agent at the local server. The agent tool calls are the same shape; only the transport changes.
Doing this yourself? MetricBridge handles the export step: all 190 metrics as local JSON, with the open-source MCP server (GitHub).
One purchase. No request count.
MetricBridge is a $24.99 one-time unlock. It covers the export, the raw MCP server, and the on-device Ask and Weekly Brief. There is no recurring plan and nothing of yours sits on our servers.