health-os
health-os is a local-first personal health record exposed over MCP, letting an MCP client read, update, and safely analyze your own health data.
View deterministic health summaries, active diagnoses, current medications, allergies, habits, and health timeline.
Query approved lab/observation values, trends, anomalies, screening recommendations, weekly reports, and doctor-visit briefs.
Track food logs, nutrients, deficiencies/excesses, meal templates, and food↔wellbeing patterns.
Run deterministic medication-safety checks, crisis-resource responses, and full-text search over narrative reports.
Use read-only SQL over approved views for custom queries and aggregates.
Record or update profile, allergies, diagnoses, medications, habits, meals, and meal templates.
Stage extracted lab panels as pending, map unknown markers, and approve panels only on explicit user confirmation.
Imports Apple Health exports, pulling wearable/device data into the local health record so it can be used in trends, baselines and analytics.
Imports Garmin device data into the local health record, feeding activity and health metrics into trends, baselines and analytics.
Delivers critical-value alerts as native macOS notifications so urgent lab findings surface immediately on the user's machine.
Optional alert channel that sends a generic "check your health system" message when critical values are detected; it never transmits actual health values.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@health-os show me my latest cholesterol lab results"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
health-os
Local-first personal health record, exposed over MCP. Your labs, diagnoses, medications, wearable data and food log live in your own Postgres; any MCP client — one running a local model or a cloud assistant — can read and update them through guarded tools. Critical values, drug-safety rules and screening schedules are deterministic code, not LLM judgement.
Medical disclaimer. This is not a medical device and does not give medical advice. Critical-value alerts and screening reminders are only a signal to contact a doctor — never a diagnosis and never a reason to delay care. Use at your own risk.
What it does
Lab results — drop a PDF or photo into your MCP client; the model extracts the values, health-os normalizes names (uk/ru/en/Latin synonyms) and units, and stages the panel as pending. Nothing counts as fact until you approve it.
Safety net in code — critical values alert immediately (log, macOS notification, optional Telegram); critical findings in narrative reports are flagged; drug-interaction questions are refused and redirected to a doctor/pharmacist (only deterministic checks run: total daily paracetamol across products, biotin before lab tests); a crisis tool returns a fixed response with hotlines, independent of the model.
Trends and analytics — Mann-Kendall trends with a per-year slope, anomalies against your own baseline (
get_anomalies), a screening calendar, a weekly report, a doctor-visit brief. Risk calculators (SCORE2, ASCVD, FRAX) so far only enforce their validated age range; the formulas aren't implemented yet.Food log — meals with a 41-nutrient profile, %RDA, deficiency/excess flags, meal templates.
Devices — Apple Health export and Garmin import.
38 MCP tools + server instructions — the safety rules are sent to every client on connect; see mcp_server/README.md.
Related MCP server: indaga-agent
Try it in one command
Only Docker needed. Starts a throwaway database with a fictional patient — two years of labs (LDL creeping up), blood pressure, medications, a food log and a lab panel awaiting approval:
git clone https://github.com/andronaft/health-os && cd health-os
make demo # or: cd demo && docker compose up -d --buildPoint your MCP client at it:
{
"mcpServers": {
"health-os-demo": {
"command": "docker",
"args": ["run", "-i", "--rm", "--network", "health-os-demo",
"-e", "DATABASE_URL=postgresql+psycopg://health:demo@db:5432/health_os",
"-e", "READONLY_DATABASE_URL=postgresql+psycopg://health_readonly:demo_ro@db:5432/health_os",
"health-os:local"]
}
}
}Ask "show my health summary", "is my LDL trending up?", "what's pending review?",
"what am I short on nutritionally?". Remove it all with make demo-down.
Install for your own data
Requires Docker and Python 3.12+.
git clone https://github.com/andronaft/health-os && cd health-os
cp .env.example .env # set the passwords
docker compose up -d db # Postgres 16 + pgvector
python3 -m venv .venv && .venv/bin/pip install -e ".[dev]"
.venv/bin/alembic upgrade head # schema
.venv/bin/python -m seed.load # marker catalog, synonyms, units, nutrients
.venv/bin/python -m seed.demo # optional: a fictional demo patient to play withThen connect an MCP client — config for LM Studio, Open WebUI, Ollama CLI and Claude is in mcp_server/README.md. Try: "show my health summary", "LDL trend", "what am I short on nutritionally this week?".
Local models: the server speaks standard MCP over stdio, so any MCP client that runs a local model can use it. Verified so far: the server itself with the official MCP Python client (CI + the Docker demo). Not yet verified end-to-end with a local model — see #9; reports welcome.
Browser UI (preview)
make ui # or: .venv/bin/python -m web [--port N] [--no-browser]Starts a local, read-only web page on 127.0.0.1:8765 (port: UI_PORT in .env) and opens it
with this run's session link. Only this machine can reach it: the server listens on 127.0.0.1,
refuses requests with a foreign Host, and its API answers only GET requests that carry the
session token. The Overview page shows your profile, what needs attention (critical values with
their escalation text, values awaiting review, latest values out of range, overdue screenings,
markers not checked in a long time), 30-day tiles for blood pressure, resting heart rate, HRV,
sleep score and weight, and recent events. It reads the same rows as get_health_summary, and
only ever says "N of M recorded markers in range". The other pages are still empty — they come in
0.4.0 (milestone). Nothing is loaded from
outside this machine: Apache ECharts is kept in web/static/vendor/ with its licence.
How it works
A lab result, from report to fact
Every row of a staged panel ends up pending — nothing becomes a fact without an explicit approve. A row the catalog can't place is kept, not dropped:
Unmapped and pending rows show up in list_pending_reviews and in the health summary's
"awaiting review" block; a learned name maps automatically on the next panel.
Directory | What's inside |
| config, DB, normalization, services, dedup, health summary |
| MCP server, read and write tools |
| deterministic safety rules and alert delivery |
| trends, baselines, calculators, screening, nutrition, reports |
| extraction schema, confidence scoring, device importers, embeddings |
| Alembic schema |
| reference catalog + the demo patient |
| red-team scenarios (injections, hidden critical values, unit tricks) |
| backup/restore (restic; |
Privacy / local-first
Your data stays in your own database. Postgres runs locally in Docker;
data/and.envare outside git. Nothing is sent anywhere by health-os itself.What leaves the machine depends on the MCP client you connect. With a local model (LM Studio, Open WebUI + Ollama, …) nothing does. With a cloud assistant, whatever the tools return is sent to that provider — use one whose terms fit medical data (no training on your data, zero/short retention).
The goal is fully local: local models for chat and extraction, local embeddings for search (already supported via fastembed). Cloud clients remain optional.
Optional alert channel (Telegram) sends only a generic "check your health system" text, never values.
Encrypt the disk (FileVault / LUKS / BitLocker) — the database files are plaintext at rest.
Never put real medical data in issues, PRs or tests — synthetic data only.
Tests
make test # everything (needs Postgres for the integration part)
make test-unit # pure unit tests — no database needed
make test-integration # only tests marked `integration`Integration tests never touch the working database: tests/conftest.py drops and recreates
<POSTGRES_DB>_test on the same server (migrations + seed) on every run. Override with
TEST_DATABASE_URL (the name must end in _test). Without Postgres, integration tests are
skipped locally; CI sets REQUIRE_DB=1 so they fail instead.
Development history: PROGRESS.md.
How the guardrails work: How to let an LLM read your medical records without trusting it. Made by Andrii Zhukovskyi.
License
AGPL-3.0-or-later. You may use, modify and fork health-os; if you distribute it or run a modified version as a network service, you must publish your source under the same license.
Want to use it in a closed-source or commercial product without those obligations? A separate commercial license is available from the author — reach out via GitHub (@andronaft).
Contributions are welcome — see CONTRIBUTING.md (includes a short CLA).
Available Tools
36 toolsapprove_staged_sourceADestructiveIdempotent
Approve a staged panel (pending→approved). ONLY on an explicit instruction from the user in the current message — do not call right after stage_lab_panel. Refuses when numeric values have an unconvertible unit (they'd be invisible to trends); allow_missing_canonical=true only if the user accepts that.
Inputs: source_id (required), allow_missing_canonical. Returns: result
check_medication_safetyARead-only
Call for ANY question about medications: "can I take X", combining drugs, doses. Returns the standard refusal to assess interactions (a doctor/pharmacist must check) plus deterministic checks: total daily paracetamol across products and biotin interference with lab tests. paracetamol_products: [{name, mg_per_dose, doses_per_day}] — include combination cold/flu remedies; if omitted, current medications are used. planned_tests: marker codes (e.g. ["tsh", "ferritin"]).
Inputs: planned_tests, taking_biotin, paracetamol_products. Returns: result
crisis_resourcesARead-only
Call IMMEDIATELY on any sign of crisis, suicidal thoughts or self-harm. Returns a fixed response with hotlines and the user's trusted contact. Reply with it as is — no analytics.
Inputs: message. Returns: result
delete_mealADestructiveIdempotent
Remove a meal from the food diary and the nutrient totals — only when the user says it was logged by mistake or twice (food_log_id from query_food). To fix a time, portion or nutrients use update_meal instead.
Inputs: food_log_id (required). Returns: result
get_allergiesARead-only
All recorded allergies: allergen, reaction, severity and whether it is verified. Unverified allergies must still be treated as real (fail-safe) — never suggest an allergen because it is "unverified". Use before any discussion of foods or medications; to add one use record_allergy.
Inputs: no named parameters. Returns: result
get_diagnosesARead-only
All recorded diagnoses with two independent status axes: clinical_status (active / resolved / …) and verification_status (confirmed / provisional / suspected / differential / refuted), plus date and ICD-10 code. Base advice only on confirmed diagnoses; present suspected/provisional ones as open questions and ignore refuted ones. Use this for "what am I diagnosed with"; get_health_summary already includes the active ones briefly.
Inputs: no named parameters. Returns: result
get_habitsARead-only
Habits the user reported — smoking, alcohol, cannabis, vaping and similar: habit_type, status (current / former / never), frequency as said (e.g. "about once a week"), optional cigarettes_per_day / smoking_years / drinks_per_week, start/end dates and notes. Former habits are listed too (pack-years and past exposure matter for risk). Take them into account for blood pressure, heart rate, lung and liver questions; state them as facts, without moralising. To add one use record_habit, to change one update_habit.
Inputs: no named parameters. Returns: result
get_health_summaryARead-only
Deterministic health summary: profile, allergies (including unverified), active diagnoses, current medications, recency of exams. A guide — exact values via query_observations.
Inputs: no named parameters. Returns: result
get_medicationsARead-only
Medications and supplements the user is currently taking (status "taking"): name, dose, unit, times per day, route, product type (prescription / otc / supplement / herbal). Use it before any medication question — then call check_medication_safety; never suggest dose changes. To add one use record_medication, to fix an entry update_medication, when the user stopped taking one stop_medication.
Inputs: no named parameters. Returns: result
get_screening_recommendationsARead-only
Screening calendar for the user's profile (age-gate, "you don't need this yet"). Statuses: due/overdue/up_to_date/not_yet. Requires a filled-in profile.
Inputs: no named parameters. Returns: result
get_timelineARead-only
Chronology of the user's health events: diagnoses, lab panels, visits, medications, hospitalizations, vaccinations — newest first, at most 200 rows of {kind, at, title}. Use it for "what happened when" questions or to orient before a visit. It lists events, not values: for numbers use query_observations, for a marker's direction use get_trend. days: how far back to look (default 3650 ≈ 10 years).
Inputs: days. Returns: result
get_trendARead-only
Statistical trend of one marker over time (Mann-Kendall test + slope) on approved values: direction increasing / decreasing / no_trend with z, p-value and slope per measurement. Needs at least 5 values in the window, otherwise returns no_trend with the reason. Use for "is X going up/down"; for the raw values use query_observations. type_code: catalog code such as ldl, hemoglobin, vitamin_d, resting_hr, hrv (as returned by query_observations / sql_query on v_observations). days: window (default 1825 ≈ 5 years).
Inputs: type_code (required), days. Returns: result
get_weekly_reportARead-only
Deterministic report for the last 7 days: newly added values, abnormal approved values, top nutrient deficiencies/excesses from the food log, and system health (review-queue size, share of corrected extractions, approved values missing a canonical value). Use for a weekly check-in; for a single marker use get_trend, for a visit use prepare_doctor_visit.
Inputs: no named parameters. Returns: result
list_meal_templatesARead-only
Saved templates for frequent meals (for quick log_from_template).
Inputs: no named parameters. Returns: result
list_pending_reviewsARead-only
Lab values that were staged (stage_lab_panel) but NOT yet approved by the user — marker, value, unit, date and status, each marked "PENDING — unverified". Never cite these as facts or use them in trends; show them so the user can compare with the original report and then approve via approve_staged_source. Rows with unmapped=true have no marker type yet (only the printed raw_name): ask the user which marker it is, then map_pending_observation. Approved values are read with query_observations.
Inputs: no named parameters. Returns: result
log_from_templateA
Log a meal from a saved template in one call: copies the template's nutrients multiplied by portion_factor (1.0 = one portion, 1.5 = one and a half). name: template name (see list_meal_templates); eaten_at: 'YYYY-MM-DD HH:MM' (default now); wellbeing / symptoms: the user's state after eating (optional). For a meal without a template use log_meal.
Inputs: name (required), eaten_at, symptoms, wellbeing, +1 more. Returns: result
log_mealA
Log a meal into the diary (auto-approved). description — the meal description (required); meal_type: breakfast/lunch/dinner/snack/drink; eaten_at=ISO 'YYYY-MM-DD HH:MM' (default — now). nutrients — {code: amount} per nutrient_types (energy_kcal/protein/carbs/fat/fiber/sugar/ added_sugar/saturated_fat/omega3/sodium/potassium/calcium/iron/magnesium/zinc/vitamin_a/ vitamin_c/vitamin_d/vitamin_b12/folate_b9/water/caffeine/alcohol/...). added_sugar — sugar ADDED to the dish (not natural from fruit/milk). The full-profile estimate is made by the model (including from a photo). glycemic_index/glycemic_load — per meal; symptoms/wellbeing — reaction after eating. Daily totals go by the date of eaten_at: when the user logs a meal later (e.g. lunch in the evening, yesterday's dinner after midnight), pass when it was eaten, not now; a time_check in the response means the time looks wrong — confirm it with the user. Review — query_food; daily norms — query_nutrition.
Inputs: description (required), notes, portion, eaten_at, +7 more. Returns: result
map_pending_observationA
Assign a marker type (observation_types.code, e.g. "lymphocytes_abs") to a pending row that is unmapped or matched to the wrong marker — only after the user confirmed which marker it is. Re-normalizes the value into the canonical unit, re-runs the critical-value check and (learn_synonym) remembers the printed name so future panels map it automatically. Refuses a type whose unit dimension doesn't fit. The row stays PENDING; approve separately.
Inputs: type_code (required), observation_id (required), learn_synonym. Returns: result
nutrition_reportARead-only
Nutrition analytics over N days: top deficiencies/excesses (%RDA+flags) + food's link to wellbeing (average GI/sugar/sodium by wellbeing category). Association, not causation.
Inputs: days. Returns: result
prepare_doctor_visitARead-only
One-call briefing to take to a doctor: the deterministic health summary, the latest abnormal approved values (high/low/critical, up to 15), screenings that are due or overdue, the number of values awaiting review, and a disclaimer. Use before an appointment or when the user asks "what should I tell my doctor". specialty: optional label echoed in the output (content is not filtered by specialty yet).
Inputs: specialty. Returns: result
query_foodARead-only
Food diary entries for the last N days, newest first (max 200 meals): id (for update_meal / delete_meal), time, meal type, description, portion, glycemic index/load, symptoms and wellbeing after eating, plus the nutrients stored for each meal. Use it to see WHAT was eaten. For daily totals vs norms use query_nutrition; for deficiencies and food↔wellbeing patterns use nutrition_report. days: look-back window (default 7). meal_type: optional filter — breakfast / lunch / dinner / snack / drink.
Inputs: days, meal_type. Returns: result
query_nutritionARead-only
Average daily intake of every logged nutrient over the last N days, with %RDA and a flag: "deficient" (<70% of the norm) or "excess" (above the safe upper limit). Averages are per LOGGED day, so incomplete logging understates intake — check days_logged. Limit-only nutrients (sugar, added sugar, saturated fat, sodium, cholesterol) are never flagged as deficient. Use for "am I getting enough X"; for a ranked summary with food↔wellbeing associations use nutrition_report; for the meals themselves use query_food. days: look-back window (default 7).
Inputs: days. Returns: result
query_observationsARead-only
Values of a marker (e.g. 'cholesterol_total') over N days. >90 days → weekly aggregation min/avg/max. Only confirmed (approved) values.
Inputs: type_code (required), days. Returns: result
record_allergyA
Add an allergy the user reports (stored immediately as approved, with manual provenance). allergen: what causes it (e.g. penicillin); reaction: e.g. rash, anaphylaxis; severity: mild / moderate / severe; verified: true only if confirmed by a doctor or test — unverified allergies are still treated as real; allergen_type: drug / food / environmental / other. Check get_allergies first to avoid duplicates.
Inputs: allergen (required), reaction, severity, verified, +1 more. Returns: result
record_diagnosisA
Add a diagnosis (stored immediately as approved, with manual provenance). diagnosis_name: as written by the doctor; diagnosed_at: YYYY-MM-DD; icd10_code: optional; clinical_status: active (default) / resolved; verification_status: confirmed (default) / provisional / suspected / differential / refuted — use suspected or provisional for anything not confirmed by a doctor; severity: optional. Check get_diagnoses first to avoid duplicates.
Inputs: diagnosed_at (required), diagnosis_name (required), severity, icd10_code, +2 more. Returns: result
record_habitA
Add a habit the user reports (stored immediately as approved): habit_type — smoking / alcohol / cannabis / vaping / other substance, lower-case; status: current (default) / former / never; frequency: as the user said it ("about once a week", "daily"); start_date / end_date: YYYY-MM-DD, only when the user gave the day — otherwise leave empty and put "since July 2026" into notes, never invent a day; notes: details (mixed with tobacco, CBD/THC, …); cigarettes_per_day / smoking_years / drinks_per_week when known. A change of pattern (e.g. stopped mixing with tobacco) = mark the old entry former via update_habit and record the new one. Check get_habits first to avoid duplicates.
Inputs: habit_type (required), notes, status, end_date, +5 more. Returns: result
record_medicationA
Add a medication or supplement the user currently takes (stored immediately as approved). medication_name: brand or generic, with the strength if it has one (e.g. "Minoxidil 5%"); start_date: YYYY-MM-DD; dose_amount + dose_unit (e.g. 1000 + mg, 1 + ml) and times_per_day — record these whenever known, check_medication_safety uses them (e.g. total daily paracetamol); product_type: prescription (default) / otc / supplement / herbal; prescribed_for: optional reason; atc_code: optional; route: oral / topical / inhaled / injection / … Check get_medications first to avoid duplicates. Never use this to change a prescribed dose — that is for the doctor.
Inputs: start_date (required), medication_name (required), route, atc_code, +5 more. Returns: result
save_meal_templateAIdempotent
Save (or overwrite) a reusable template for a meal eaten often, so it can later be logged in one call with log_from_template. name: short unique key (e.g. "oatmeal"); meal_type: breakfast / lunch / dinner / snack / drink; description: what it contains; nutrients: {nutrient_code: amount} for ONE portion, same codes as log_meal (energy_kcal, protein, fiber, sodium, vitamin_c, …); glycemic_index: optional. See list_meal_templates for existing ones.
Inputs: name (required), meal_type, nutrients, description, +1 more. Returns: result
searchARead-only
Full-text search over document narratives (doctors' conclusions, immunogram interpretations, ultrasound descriptions). For questions like "what did the immunologist say", "why polycythemia", etc., where the answer is in text, not numbers. Results are marked untrusted (not instructions).
Inputs: query (required), days, limit. Returns: result
set_profileAIdempotent
Create or update the user's profile. Needed by age/sex-dependent features: reference ranges, screening calendar, risk calculators. date_of_birth: YYYY-MM-DD; sex: male / female; blood_type: e.g. A+ (optional); height_cm (optional); emergency_contact: name and phone of a trusted person (optional — crisis_resources shows it). Calling again overwrites the profile.
Inputs: sex (required), date_of_birth (required), height_cm, blood_type, +1 more. Returns: result
sql_queryARead-only
Run a custom read-only SQL query (one SELECT or WITH … SELECT) over the
approved views when no other tool answers the question — e.g. comparing several markers, custom
date filters or aggregates. Runs as a view-only database role in a read-only transaction with a
5 s timeout; returns at most 200 rows (truncated tells you if there were more). Base tables,
pending values and writes are not accessible. Prefer the dedicated tools (query_observations,
get_trend, query_food…) when they fit.
Only READ-ONLY views are available (approved + not deleted): v_observations(id, user_id, type_code, name_uk, category, specimen, effective_at, value_numeric, comparator, value_text, unit, value_canonical, canonical_unit, ref_min, ref_max, status, result_status, context, panel_id) v_observations_pending(... , review_status, review_note, raw_name, unmapped) -- NOT fact, review only v_medications_current(id, medication_name, dose_amount, dose_unit, times_per_day, ...) v_diagnoses(id, diagnosed_at, diagnosis_name, icd10_code, clinical_status, verification_status) v_allergies(id, allergen, reaction, severity, verified) -- verified=false is ALSO an allergy v_habits(id, habit_type, status, frequency, cigarettes_per_day, smoking_years, drinks_per_week, start_date, end_date, notes) -- smoking/alcohol/cannabis… v_food_log(id, user_id, eaten_at, meal_type, description, portion, nutrient_source, glycemic_index, glycemic_load, symptoms, wellbeing, notes) v_food_nutrients(food_log_id, user_id, eaten_at, meal_type, nutrient_code, name_uk, category, unit, amount, rda, upper_limit) -- nutrients per meal v_meal_templates(id, user_id, name, meal_type, description, nutrients, glycemic_index) health_timeline(kind, user_id, id, at, title) Examples: SELECT type_code, effective_at, value_canonical FROM v_observations WHERE type_code='cholesterol_total' ORDER BY effective_at DESC; SELECT * FROM v_diagnoses WHERE verification_status='confirmed';
Inputs: sql (required). Returns: result
stage_lab_panelA
Stage an extracted lab panel (PENDING). panel_date: YYYY-MM-DD from the report. rows: a list of {raw_name, value, unit, ref_min, ref_max}; value is a number — a printed "6,4" is read as 6.4, and a qualitative result ("negative", "not detected") or a limit ("<0.5") goes in as text (or pass it as value_text). Critical values are alerted immediately. Names not in the catalog (or whose unit can't belong to the matched marker) are stored UNMAPPED and pending — resolve them with map_pending_observation. Afterwards — show the table to the user and wait for an explicit approve_staged_source.
Inputs: rows (required), panel_date (required), facility, panel_type. Returns: result
stop_medicationAIdempotent
Record that the user no longer takes a medication: medication_id from get_medications, end_date YYYY-MM-DD (default today; not in the future), stop_reason and side_effects optional (e.g. "course finished", "doctor stopped it", "dose changed"; "rash"). It drops out of get_medications and the safety checks but stays on the timeline as medication_stop. Only when the user says it was stopped — never suggest stopping a medication. If the side effect sounds like an allergic reaction, offer record_allergy too.
Inputs: medication_id (required), end_date, stop_reason, side_effects. Returns: result
update_habitAIdempotent
Correct a recorded habit or record a change in it: habit_id from get_habits; pass only the fields to change (empty = keep). "I quit in September" = status former + end_date; "now twice a week" = frequency. notes replaces the old text — pass the full new note. Dates YYYY-MM-DD, not in the future.
Inputs: habit_id (required), notes, status, end_date, +6 more. Returns: result
update_mealAIdempotent
Correct a logged meal: food_log_id from query_food; pass only what to change (empty = keep). eaten_at 'YYYY-MM-DD HH:MM' moves it to the right day (daily totals go by its date); nutrients {code: amount} overwrites just those codes, the rest of the profile stays — e.g. the user ate half: pass the halved amounts. To remove a meal logged by mistake or twice use delete_meal.
Inputs: food_log_id (required), notes, portion, eaten_at, +7 more. Returns: result
update_medicationAIdempotent
Correct a recorded medication the user says is wrong or incomplete: a typo in the name, a missing or mistyped dose, unit, times_per_day or route, the wrong start date. medication_id comes from get_medications; pass only the fields to change (empty = keep). The previous version is kept in the change history. Not for a dose the doctor changed — then call stop_medication (stop_reason "dose changed") and record_medication with the new dose from that date, so the timeline shows the change. Never suggest dose changes yourself.
Inputs: medication_id (required), route, atc_code, dose_unit, +6 more. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 36 tools
Most tools have clearly distinct clinical, nutritional, medication, lab, and habit scopes, and descriptions explicitly route between similar-looking tools such as query_food, query_nutrition, and nutrition_report. Some overlap remains among summary/report tools and the generic sql_query versus dedicated read tools, but an agent can usually select correctly.
The set overwhelmingly uses consistent snake_case names, mostly with recognizable verb_noun patterns such as get_medications, record_habit, update_meal, and approve_staged_source. Minor deviations include noun-first names like nutrition_report, sql_query, and crisis_resources, but the convention is still predictable.
With 36 tools, the set is large enough to impose significant selection overhead on an agent even for a broad health domain. Several read/report/summary tools could likely be consolidated or made more task-oriented without losing coverage.
Core workflows for meals, medications, habits, lab review, and reporting are well covered, but lifecycle gaps remain: allergies and diagnoses have create/read tools but no update/delete/correction tools. Visits, vaccinations, and hospitalizations appear in the timeline but have no obvious recording tools, which are notable omissions for a health OS.
Maintenance
Related MCP Connectors
Private-by-default, local-first memory/context/task orchestrator for MCP apps and agents.
Query your own knowledge graph from any MCP client. The graph lives in a database you own.
Governed personal world model and memory for your AI agent. Pair once, connect over MCP.
Hosted Apple Health MCP for Claude and ChatGPT, works outside the US. iPhone app, no local server.
Related MCP Servers
- AlicenseBqualityBmaintenanceA local-first MCP server that enables AI agents to read user-authorized Google Health API v4 data from Fitbit, Pixel Watch, and partners via OAuth, with tokens never leaving the machine.26875 npm84MIT
- AlicenseNot gradedqualityBmaintenanceA local-first MCP server for querying multi-omic personal health data (genome, labs, wearables) with an honesty contract and progressive disclosure skills.1AGPL 3.0
- AlicenseCqualityDmaintenanceA local-first, model-agnostic MCP server that stores personal health data in a SQLite file and provides analysis-ready views for any AI client to log, retrieve, and reason over health records.79MIT
- AlicenseNot gradedqualityCmaintenanceEnables AI clients to securely access, query, and mutate normalized user-controlled health data (e.g., from Apple Health or Supabase) through a bounded set of MCP tools, with optional OAuth and sandboxed deployment.MIT