Skip to main content

Local MCP server for Vaultbeat — AI Health Sync: your AI agent reads your end-to-end-encrypted Apple Health data (sleep, cycle, weight, water), decrypted only on your machine

Project description

Vaultbeat MCP Server

Let your own AI agent read your health data — without the cloud ever seeing it.

This is the official local MCP server for Vaultbeat — AI Health Sync (formerly named Tether), the iOS app that syncs Apple Health data (sleep, heart rate, menstrual cycle, weight, water, symptoms) between partners and to your own AI — end-to-end encrypted.

Vaultbeat embeds no AI and runs no model on your phone. Intelligence lives where you control it: Claude Code, Claude Desktop, or any MCP-capable agent running on your own machine. This server is the bridge — it holds a private key that never leaves your computer, pulls ciphertext from the cloud, and decrypts only locally.

iPhone (Apple Health) ──E2EE──▶ cloud (ciphertext only) ──E2EE──▶ this server (your machine) ──▶ your AI agent

Requirements

  • Vaultbeat — AI Health Sync on iOS, with a Pro subscription (the AI-agent interface is the Pro tier)
  • Python 3.11+ on the machine where your agent runs (macOS / Linux / Windows)

Quick start

1. Install

With uv (recommended — no clone needed):

uvx vaultbeat-mcp status

Or with pip:

pip install 'vaultbeat-mcp[qr]'

Upgrading from the old tether-mcp package? Same code, new name — your existing binding and config carry over unchanged. Just swap the package name in your install command and MCP client config.

2. Bind your phone

vaultbeat-mcp bind

This generates a keypair on your machine and prints a QR code. In the Vaultbeat iOS app, open Settings → Data & AI → MCP Server and scan it (or import a QR screenshot from Photos). The app authorizes this machine and starts sealing your health envelopes to its public key. The private key stays in ~/.tether/mcp-local/ (owner-only 0600 permissions, OS keychain where available) — it is never uploaded anywhere. (The directory keeps its original pre-rename path so existing bindings survive upgrades.)

3. Connect your agent

Claude Code (one line):

claude mcp add vaultbeat-health -- uvx vaultbeat-mcp serve --transport stdio

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "vaultbeat-health": {
      "command": "uvx",
      "args": ["vaultbeat-mcp", "serve", "--transport", "stdio"]
    }
  }
}

Any other MCP client: run vaultbeat-mcp serve --transport stdio, or serve --transport http for a loopback streamable-HTTP endpoint with bearer-token auth.

Claude Desktop note: it does not inherit your shell PATH. If uvx isn't found, use the absolute path (which uvx) as command.

Debugging: npx @modelcontextprotocol/inspector uvx vaultbeat-mcp serve --transport stdio

Then just ask your agent: “How did we sleep last night?”

MCP tools (16)

Tool Returns
vaultbeat_status Local binding state (never exposes keys or tokens)
vaultbeat_doctor Full self-diagnosis: install/binding chain plus which data types have data and which need a newer iOS build — call this before concluding data is missing
vaultbeat_start_binding A fresh QR binding payload for the iOS app to scan
vaultbeat_poll_binding One poll for the iOS authorization to complete binding
vaultbeat_sync_sleep Recent sleep sessions incl. heart-rate samples, per-day primary-session selection matching the iOS app
get_sleep_detail Per-night heart-rate + respiratory-rate + sleep-stage timeline
get_water_intake Daily water intake + computed daily average
get_weight_trend Daily weights + latest/avg/min/max + weekly trend rate
get_menstrual_cycle Cycle samples + next-period prediction (sensitive — explicit iOS opt-in required)
get_symptoms HealthKit symptom days grouped by data owner (sensitive)
get_notes Free-text day annotations with their writer (sensitive)
get_activity Daily activity rings: steps / energy / exercise minutes / stand hours / distance
get_resting_hr Resting heart-rate records + window mean
get_workouts Workout records: type / duration / calories / distance
get_mindfulness Mindful sessions and minutes per day
get_hrv Heart-rate variability (SDNN) records + window mean
get_wrist_temp Sleeping wrist-temperature baseline deviation
get_hrv_hourly (via get_hrv(granularity="hourly")) Hour-bucketed HRV averages over 30 days — the context-cheap default; granularity="raw" keeps minute-level spike precision
get_vo2max VO₂ max records + latest / peak / trough / window average
get_basal_energy Basal (resting) energy burned, per hour bucket
get_total_energy_burned TDEE — basal + active per day, measured rather than estimated, with today flagged partial and excluded from the average
get_strength_log Structured strength training: exercise, sets, reps, weight per day
get_food_log Meals as free text with optional portions and timing
log_weight_entry Write a weight entry (optionally mirrored into Apple Health when the user opts in)
log_strength_entry Write a strength-training entry for a given day
log_food_entry Write a meal entry for a given day
log_note Write a mood or general note for a given day

Every data tool accepts owner (a user-ID prefix) to filter to one person — the server may hold both your and your partner's shared records, and omitting owner mixes them into one pool, so per-person questions should always pass it. (Earlier releases named some tools get_partner_* / tether_*; they were renamed in the 16-tool and Vaultbeat releases.)

Reads are cache-first: decrypted records are cached locally (owner-only files, 600 s TTL, VAULTBEAT_MCP_CACHE_TTL to override — the pre-rename TETHER_MCP_* spellings still work) so repeat queries answer in ~0.2 s with zero network; pass fresh=true to force a cloud round trip. The same service layer backs a full CLI (vaultbeat-mcp sleep / water / weight / … — every data subcommand takes --owner too) if you prefer scripts over MCP.

Privacy & security model

  • End-to-end encryption: Curve25519 ECDH + HKDF-SHA256 + AES-GCM. Every health record is sealed on-device to each authorized recipient's public key (your partner, and this server once bound).
  • The cloud only ever holds ciphertext. Vaultbeat's backend cannot read your health data — architecturally, not just by policy.
  • Decryption happens here, on hardware you own. The private key and server token are never exposed through any tool result.
  • Sensitive kinds (menstrual cycle, symptoms, notes) reach this server only if explicitly opted in inside the iOS app, and are never re-exported by the server.
  • HTTP transport binds to loopback by default and requires a bearer token; binding a non-loopback address fails closed unless explicitly allowed — front it with TLS if you must expose it.

You can audit all of the above in this repository — that is why it is open source.

Development

pip install -e '.[dev,qr]'
pytest

License

MIT. The Vaultbeat iOS app and cloud service are separate proprietary components; this repository covers the local MCP server only.


Website: vaultbeat.app · App Store: Vaultbeat — AI Health Sync · Bugs & feedback: vaultbeat-community

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

vaultbeat_mcp-0.2.3.tar.gz (97.8 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

vaultbeat_mcp-0.2.3-py3-none-any.whl (72.5 kB view details)

Uploaded Python 3

File details

Details for the file vaultbeat_mcp-0.2.3.tar.gz.

File metadata

  • Download URL: vaultbeat_mcp-0.2.3.tar.gz
  • Upload date:
  • Size: 97.8 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.10.2 {"installer":{"name":"uv","version":"0.10.2","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for vaultbeat_mcp-0.2.3.tar.gz
Algorithm Hash digest
SHA256 163106cd3657f50c3b844e8a44e882acc762f02b6a9b6a969cf46b58063ed566
MD5 af793b5b828993bc744fe75a4e1b1e44
BLAKE2b-256 6039ceda7e042453b28738fa7a8d431a457ce68bae257448e430dda54d589c39

See more details on using hashes here.

File details

Details for the file vaultbeat_mcp-0.2.3-py3-none-any.whl.

File metadata

  • Download URL: vaultbeat_mcp-0.2.3-py3-none-any.whl
  • Upload date:
  • Size: 72.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.10.2 {"installer":{"name":"uv","version":"0.10.2","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for vaultbeat_mcp-0.2.3-py3-none-any.whl
Algorithm Hash digest
SHA256 0a2ca0de2e2a49bdc7987148b530e1da23611ddd54761554ddeb5562d920499e
MD5 7563e29a70d278e14c5be97636a116c8
BLAKE2b-256 00b146de0c9b25a2b2febfb8792ce7f6786d03754c04e4ae1059b087be65c771

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page