Wrong number of steps and general approach

[bartmichalak]

So I asked Claude to analyse my daily average of steps in September:

Based on my Apple Health data, could you make a summary of my daily average steps in September?

{
  `params`: {
    `limit`: 1000,
    `date_to`: `2025-09-12T23:59:59+00:00`,
    `date_from`: `2025-09-01T00:00:00+00:00`,
    `record_type`: `HKQuantityTypeIdentifierStepCount`
  }
}

so then Claude used Get health records:

{
  `params`: {
    `limit`: 1000,
    `date_to`: `2025-09-12T23:59:59+00:00`,
    `date_from`: `2025-09-01T00:00:00+00:00`,
    `record_type`: `HKQuantityTypeIdentifierStepCount`
  }
}

and the response was like:

[{"type":"HKQuantityTypeIdentifierStepCount","sourceVersion":"11.5","sourceName":"Bartosz’s Apple Watch","device":"<<HKDevice: 0x739e67ac0>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","startDate":"2025-09-01T03:50:26Z","endDate":"2025-09-01T03:51:13Z","creationDate":"2025-09-01T03:52:38Z","unit":"count","value":19.0,"textvalue":"19"},{"type":"HKQuantityTypeIdentifierStepCount","sourceVersion":"11.5","sourceName":"Bartosz’s Apple Watch","device":"<<HKDevice: 0x739e67ac0>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","startDate":"2025-09-01T03:53:09Z","endDate":"2025-09-01T03:53:22Z","creationDate":"2025-09-01T04:01:01Z","unit":"count","value":24.0,"textvalue":"24"},

Unfortunately, I didn't receive the full response, because:

Result too long, truncated to 100000 characters

It's due to the fact, that only from the September, I had 2626 samples of data, and I think it makes no sense to process it by LLM, given the fact it costs tokens.

After that try, Claude used another tool Get trend data:

{
  `date_to`: `2025-09-12T23:59:59+00:00`,
  `interval`: `day`,
  `date_from`: `2025-09-01T00:00:00+00:00`,
  `record_type`: `HKQuantityTypeIdentifierStepCount`
}

Suggestion: Maybe the tool description should give explicit instructions that this tool shouldn't be used for analysis longer than one day or so?

and the response was like:

[{"interval":"2025-09-01T00:00:00Z","average":24.81578947368421,"min":1.0,"max":1787.0,"count":836},{"interval":"2025-09-02T00:00:00Z","average":24.25,"min":1.0,"max":280.0,"count":172},{"interval":"2025-09-03T00:00:00Z","average":151.0990990990991,"min":1.0,"max":1087.0,"count":111},{"interval":"2025-09-04T00:00:00Z","average":114.64367816091954,"min":2.0,"max":888.0,"count":87},{"interval":"2025-09-05T00:00:00Z","average":106.6896551724138,"min":1.0,"max":1064.0,"count":87},{"interval":"2025-09-06T00:00:00Z","average":121.08988764044943,"min":2.0,"max":582.0,"count":89},{"interval":"2025-09-07T00:00:00Z","average":127.0,"min":2.0,"max":768.0,"count":78},{"interval":"2025-09-08T00:00:00Z","average":22.849598163031,"min":1.0,"max":1789.0,"count":871},{"interval":"2025-09-09T00:00:00Z","average":103.96446700507614,"min":1.0,"max":1085.0,"count":197},{"interval":"2025-09-10T00:00:00Z","average":139.96703296703296,"min":3.0,"max":1088.0,"count":91},{"interval":"2025-09-11T00:00:00Z","average":50.57142857142857,"min":5.0,"max":102.0,"count":7}]

It's better, but since there's no sum, Claude had to calculate the daily value separately using a Python script that it created for this purpose. I think this unnecessary additional step should be handled by the MCP server by adding a sum field.

At the end I received these values:

Sept  1: 20,750 steps (836 records)
Sept  2: 4,171 steps (172 records)
Sept  3: 16,772 steps (111 records)
Sept  4: 9,974 steps (87 records)
Sept  5: 9,282 steps (87 records)
Sept  6: 10,777 steps (89 records)
Sept  7: 9,906 steps (78 records)
Sept  8: 19,902 steps (871 records)
Sept  9: 20,480 steps (197 records)
Sept 10: 12,737 steps (91 records)

In the end, I got these values which aren't correct - I'm attaching the actual values (from Apple Health app) below:

Sept  1: 11584
Sept  2: 3868 
Sept  3: 10049
Sept  4: 4654
Sept  5: 5982
Sept  6: 6354
Sept  7: 5754
Sept  8: 10923
Sept  9: 11658
Sept 10: 6850

So as you can see, all the values look doubled. I think it's because steps are counted separately by the Watch and phone. Maybe the MCP server should look at the data and set up a default priority for values from the watch?

[czajkub]

As for the truncating of longer responses, the bigger underlying issue appears to be that the tools return such large responses - maybe there could be a max limit implemented for the search of records, although with the addition of sums to the tools it (hopefully) shouldn't have a reason to ask for such a large sample of data

[bartmichalak]

I've tested the newest changes, but unfortunately I'm still experiencing some issues. I can't get an answer to the question: Give me my step total for each day in September so far

When I ask this question, Claude uses Get trend data duckdb tool:

{
  `date_to`: `2025-09-23T23:59:59+00:00`,
  `interval`: `day`,
  `date_from`: `2025-09-01T00:00:00+00:00`,
  `record_type`: `HKQuantityTypeIdentifierStepCount`
}

and this is the fragment of the answer:

[{"device":"<<HKDevice: 0x77132b000>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":25.0,"sum":175.0,"min":8.0,"max":76.0,"count":7},{"device":"<<HKDevice: 0x77132aa00>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":7.251256281407035,"sum":1443.0,"min":3.0,"max":9.0,"count":199},{"device":"<<HKDevice: 0x771329020>, name:iPhone, manufacturer:Apple Inc., model:iPhone, hardware:iPhone15,2, software:18.5, creation date:2025-06-28 13:36:16 +0000>","interval":"2025-09-01T00:00:00Z","average":385.5,"sum":771.0,"min":4.0,"max":767.0,"count":2},{"device":"<<HKDevice: 0x7713290e0>, name:iPhone, manufacturer:Apple Inc., model:iPhone, hardware:iPhone15,2, software:18.5, creation date:2025-06-28 13:36:16 +0000>","interval":"2025-09-01T00:00:00Z","average":346.42105263157896,"sum":6582.0,"min":1.0,"max":1787.0,"count":19},{"device":"<<HKDevice: 0x771329500>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":7.26,"sum":1452.0,"min":3.0,"max":9.0,"count":200},{"device":"<<HKDevice: 0x771328660>, name:iPhone, manufacturer:Apple Inc., model:iPhone, hardware:iPhone15,2, software:18.5, creation date:2025-06-28 13:36:16 +0000>","interval":"2025-09-01T00:00:00Z","average":25.5,"sum":51.0,"min":25.0,"max":26.0,"count":2},{"device":"<<HKDevice: 0x771328360>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":34.40331491712707,"sum":6227.0,"min":1.0,"max":943.0,"count":181},{"device":"<<HKDevice: 0x771328f60>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":7.44,"sum":1488.0,"min":2.0,"max":9.0,"count":200},{"device":"<<HKDevice: 0x77132b4e0>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-01T00:00:00Z","average":31.92,"sum":798.0,"min":1.0,"max":146.0,"count":25},{"device":"<<HKDevice: 0x77132b3c0>, name:iPhone, manufacturer:Apple Inc., model:iPhone, hardware:iPhone15,2, software:18.5, creation date:2025-06-28 13:36:16 +0000>","interval":"2025-09-01T00:00:00Z","average":1759.0,"sum":1759.0,"min":1759.0,"max":1759.0,"count":1},{"device":"<<HKDevice: 0x77132b4e0>, name:Apple Watch, manufacturer:Apple Inc., model:Watch, hardware:Watch7,9, software:11.5, creation date:2025-05-26 23:00:07 +0000>","interval":"2025-09-02T00:00:00Z","average":23.73006134969325,"sum":3868.0,"min":1.0,"max":280.0,"count":163},{"device":"<<HKDevice: 0x771329020>, name:iPhone, manufacturer:Apple Inc., model:iPhone, hardware:iPhone15,2, software:18.5, creation date:2025-06-28 13:36:16 +0000>","interval":"2025-09-02T00:00:00Z","average":33.666666666666664,"sum":303.0,"min":11.0,"max":88.0,"count":9},

As you can see, there are lots of entries with the interval date of 2025-09-01. Previously, when I tested this type of question, the intervals were not duplicated.

I recently upgraded to the newest version of iOS – could that have caused an issue? Have you tried running the latest version on the sample dataset linked in the repository?

[czajkub]

I think i see what the issue is - the device ID seems to be changing multiple times in a day (e.g. "device":"<<HKDevice: 0x77132b000>, "device":"<<HKDevice: 0x77132aa00>). I have no idea whether this is intended from Apple or something that only happens after a specific update / to specific devices, but I think I see the solution.

A better option would be to group entries by model (Iphone, Watch) instead of device id (<<HKDevice: 0x77132aa00>), which would also solve the issue of having multiple data sources listed when upgrading your iOS version for example.