io.github.davidmosiah/apple-health-mcp

Machine-readable install, runtime and privacy guidance for AI agents operating Apple Health export data.

Explain supported Apple Health export data, unavailable live HealthKit access, privacy modes and recommended agent workflow.

Check local Apple Health export path, Node version, privacy mode and Hermes client posture without reading full export data.

Personalized 3-step setup walkthrough for the human user. Adapts to current state (is APPLE_HEALTH_EXPORT_PATH set? does the export file exist and parse?). Call this first when the user asks 'how do I connect Apple Health?'. This connector is local-first and never touches Apple servers or cloud APIs.

Returns realistic example payloads of apple_health_daily_summary, apple_health_weekly_summary, and apple_health_wellness_context with Apple-Watch-style values, so agents see the contract before parsing a real export.

Return the local privacy and export-file posture without revealing health data.

List bounded records from a local Apple Health export.xml. Use type/start/end filters to keep output small. `limit` caps the returned LIST only: in summary privacy mode (the default) the `aggregate` block (`count_by_type`, `units`, `date_range`, `numeric`) is computed over every record matching the filter, and `truncated`/`limit_applied`/`matched_count` say whether the list itself was cut. The statistics are nested under `numeric` (numeric.count/sum/average/min/max), not at the top of `aggregate`. COST: because the aggregate must cover every match, summary mode reads the whole export file on each new query — roughly 33 ms per MB of export.xml (~3 s for 84 MB, ~11 s for 336 MB), and type/start/end do not shorten it. Repeating an identical query is served from an in-memory cache and returns instantly. For a cheap bounded page with no full-file scan, pass privacy_mode 'structured' or 'raw', which stop at `limit`.

List bounded workouts from a local Apple Health export.xml. `limit` caps the returned LIST only: in summary privacy mode (the default) the `aggregate` block (`count_by_activity`, `date_range`, `total_duration_minutes`, `total_distance`, `distance_units`, `total_energy_kcal`, `workout_count`) is computed over every workout matching the filter, and `truncated`/`limit_applied`/`matched_count` say whether the list itself was cut. COST: this call reads the whole export file on each new query — roughly 33 ms per MB of export.xml (~3 s for 84 MB, ~11 s for 336 MB) — and start/end do not shorten it. Workouts are sparse in an export, so even non-summary modes usually reach the end of the file. Repeating an identical query is served from an in-memory cache and returns instantly.

Check how recently the local Apple Health export file/directory was written. Returns mtime, days_since_export, an is_stale flag, and a recommendation. Considered stale if the export is older than 30 days, or older than 7 days with no recent records (the inventory's latest-record date is also older than 7 days). Use before relying on apple_health_daily_summary or apple_health_wellness_context to confirm the export is fresh.

Scan the local Apple Health export once and report available record types, workouts, date coverage, freshness and safe next calls.

Build a daily wellness summary from local Apple Health export data. It is not live HealthKit and not medical advice.

Normalize local Apple Health export sleep, workout and activity data into the shared wellness_context shape for recommendation engines.

Build a weekly wellness summary from local Apple Health export data. It is not live HealthKit and not medical advice.

Read the shared Delx Wellness profile from ~/.delx-wellness/profile.json. Returns preferred name, goals, devices, training/nutrition/exercise/agent preferences and safety flags. NEVER contains OAuth tokens or API secrets — this connector is local-export and has no cloud auth, but the profile contract is the same across every Delx Wellness MCP. Read-only.

Persist a partial patch to ~/.delx-wellness/profile.json. Requires explicit_user_intent=true (otherwise returns USER_ACTION_REQUIRED). Rejects secret-like fields (oauth, token, secret, password, cookie, refresh, api_key, session) at write time. Use to record preferred name, goals, devices, training context, nutrition context, exercise preferences, agent preferences, and safety flags.

Manually clear the incremental import cache at ~/.apple-health-mcp/incremental-cache.json. The cache tracks the latest parsed timestamp per HealthKit category so subsequent `apple_health_list_records` calls (with `incremental_cache: true`) skip already-seen records. Use this when you want to force a full re-parse without changing the export file. The cache also auto-invalidates when the export file mtime changes. Gated by explicit_user_intent: true (requires explicit user intent to wipe local cache).

Re-scan the configured watch folder (APPLE_HEALTH_WATCH_PATH or `setup --watch-path <dir>`) for a newer Apple Health export. If a newer export.xml/export.zip/apple_health_export directory is found, it is promoted to the active export, the in-memory snapshot cache and incremental cache are cleared, and subsequent summaries reflect the new data. With check_only=true, only report what would happen without promoting. This is the cross-platform recurring-refresh path — the native HealthKit bridge needs macOS and is separate. Requires explicit user intent when promoting a new export (force or non-check_only); check_only is read-only inspection.

Return the 11-question onboarding flow plus the current profile state and missing fields. Read-only — does NOT persist anything. Pair with apple_health_profile_update once the user answers. Cross-connector: the same profile is shared by every Delx Wellness MCP (whoop, garmin, oura, fitbit, strava, polar, withings, apple-health, samsung-health, google-health, nourish, cycle-coach, cgm, air).