About healthfolder
healthfolder is a local MCP server and CLI for people using Apple Watch data exported through Health Access. It reads workouts and health records from local files and can write workout ratings and training plans back without sending data over a network.
Key features
- 01Read and filter workout records
- 02Review vitals and weekly training summaries
- 03Store athlete effort ratings
- 04Validate and write training plans
- 05Inspect day files, catalog, profile, and ECGs
Use cases
- 01Review recent workouts by sport
- 02Track vitals and training volume over time
- 03Record perceived workout effort
- 04Prepare a training plan for the Watch
MCP support
Exposes workout, vitals, training, plan, and health-file tools to MCP clients over local stdio. It also provides an index resource and a workout resource template.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
13Saved tool list · updated 1w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| limit | integer | No | Maximum rows to return (default 20) |
| since | string | No | Only workouts starting on or after this date (YYYY-MM-DD) or RFC 3339 timestamp |
| sport | string | No | Activity filter: pool_swim, indoor_row, cycle, walk, run, ... (swim matches both swim kinds) |
Details
From the project READMEhealthfolder
Read a Health Access folder of Apple Watch workouts and vitals, and write training plans back into it — as plain files. The folder an app writes from Apple Health (workouts, day files, an index, a profile) is the whole API: this package is a local stdio MCP server plus a CLI over that folder. No account, no network, nothing leaves the machine.
How the folder is found
First hit wins:
--dir <folder>on the command line- the
HEALTHACCESS_DIRenvironment variable (aliasTIDE_DIR) dir = "..."in~/.config/healthaccess/config.toml(the path can be overridden withHEALTHACCESS_CONFIG; zone bounds can also be set there aszones = [133, 146, 158, 169])- discovery: iCloud Drive, then
~/Library/CloudStorage/*,~/Dropbox,~/Google Drive,~/OneDrive*,~/Box, then~/Documents. A Health Access folder holds anindex.jsonof schemahealthaccess.index.v1and aworkouts/directory. If discovery finds several, the server refuses to guess and lists them.
Run healthfolder --check to print the resolved folder, where it came
from, and the zone bounds in effect.
CLI
npx -y @reneturcios/healthfolder --check
npx -y @reneturcios/healthfolder list --sport swim --limit 5
npx -y @reneturcios/healthfolder show 7A1C3E90
npx -y @reneturcios/healthfolder vitals --days 14
npx -y @reneturcios/healthfolder summary --weeks 4
npx -y @reneturcios/healthfolder rate 7A1C3E90 7 "legs heavy, breathing fine"
npx -y @reneturcios/healthfolder plan push ./s30.json
npx -y @reneturcios/healthfolder plan status 2026-10-24T20:22:05Z-s30
npx -y @reneturcios/healthfolder plan list
npx -y @reneturcios/healthfolder day 2026-05-14
npx -y @reneturcios/healthfolder days --from 2026-05-01 --to 2026-05-14 --groups heart,sleep
npx -y @reneturcios/healthfolder catalog
npx -y @reneturcios/healthfolder profile
npx -y @reneturcios/healthfolder ecgs
Every command prints pretty JSON. Exit codes: 0 success, 1 domain or
IO error, 2 usage error. --dir works on every subcommand.
MCP server
npx -y @reneturcios/healthfolder mcp
MCP client config:
{
"mcpServers": {
"healthfolder": {
"command": "npx",
"args": ["-y", "@reneturcios/healthfolder", "mcp"]
}
}
}
Add "--dir", "<absolute path>" to args to skip the folder search.
Tools (13)
| Tool | What it does |
|---|---|
list_workouts | Workouts from the index (deduplicated, newest first), one summary line each |
get_workout | One workout in full: document, summary, parts text, zones, cardiac cost |
get_vitals | Resting HR, HRV, body mass, sleep, steps, VO2max, blood pressure over the last N days |
training_summary | One entry per ISO week: volume, zones, effort, cardiac cost, rings from day files |
rate_workout | Store the athlete's own effort rating (1–10) in the workout file |
push_plan | Validate a training plan and write plans/<id>.json for the app to put on the Watch |
get_plan_status | A plan's state: pending, scheduled, failed or cancelled |
list_plans | Every plan with its state and planned distance/time |
get_day | One day file with every Health group it holds |
get_days | Day files in a range (max 366 days), optionally reduced to named groups |
get_catalog | catalog.json plus the export-coverage check |
get_profile | profile.json as written |
list_ecgs | Every ECG summary in ecg/, newest first |
All results are pretty-printed JSON text with floats rounded to one
decimal. Malformed arguments are JSON-RPC InvalidParams errors;
domain problems (unknown id, failed validation, a refusal) come back
as isError results with the reason in plain words.
Two safety rules are built in:
rate_workoutonly ever stores a number the athlete said. A workout that already carries Apple's own rating (effort_source: "apple") is refused unless the call passesoverride: true— the refusal names the file, the existing rating, and the remedy, and nothing is written.push_planreports every validation violation in one message (sport, name length, block/step shape, goal type and value, zone, label length), so a plan can be fixed in one pass.
Resources: healthaccess://index (the deduplicated index) and the
template healthaccess://workout/{id} (one workout, heart-rate series
included). MCP prompts are not implemented in this build.
Derived numbers
- Zones. Four bounds, five zones. Bounds come from config.toml,
else are estimated from the age in
profile.json: max HR = 208 − 0.7 × age, bounds at 72/79/86/92 % of it (age 34 → [133, 146, 158, 169]). Zone seconds are computed from the heart-rate series: each sample's time until the next sample goes to that sample's zone, only time inside the workout counts, gaps over 60 s count as 60 s. With no bounds known, a file's own legacy zones are shown, markedbounds_source: "file". - Cardiac cost (swims only) = avg HR × best freestyle 100 m pace (s) / 60 = heart beats per 100 m; lower is better. The best 100 is the fastest work step of at least 95 m whose overlapping laps were at least half freestyle, normalised to 100 m; if no step qualifies and the swim is at least 100 m, the whole-workout pace is used.
- Summary lines look like:
2026-05-14 Pool swim "S12" · 1200 m · 36:06 · avg HR 129 (max 162) · Z1–2 93% · effort 6/10 (apple) · best FR 100 2:12 → 284 beats/100 m— each segment appears only when its data exists.
Presence handshake
On initialize, and after tool calls at most once a minute, the server
writes mcp/clients/<id>.json in the folder so the app can show
"Connected". The id is the FNV-1a 64 hash of machine name + client
name; the file carries the client name/version from the initialize
request, the machine name (HEALTHACCESS_MACHINE overrides it), call
counts carried over between writes, and the last tool name — never
arguments or results. A failed write never fails a tool call. The CLI
writes the same file, as client healthfolder-cli, only from the
commands that write into the folder anyway (rate, plan push).
Docker
docker run -i ghcr.io/cloudcorprecords/healthfolder # stdio MCP server
docker run --rm ghcr.io/cloudcorprecords/healthfolder list --dir /data
Mount the Health Access folder into the container and pass --dir
(or set HEALTHACCESS_DIR) so the server can find it.
Development
npm install
npm run build # tsc -> dist/
npm test # tsx --test test/*.test.ts (synthetic folders, no network)
npm run dev # run the CLI from source
License
MIT — Copyright (c) 2026 Rene Turcios.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.