Renpho Health MCP

Own your Renpho smart-scale data — pull it out of the Renpho app into a raw time series log on your machine, and explore it conversationally from an AI agent.

Renpho scales only expose their readings through the Renpho mobile app — no export, no official API. This project logs in with your Renpho account, pulls every reading (weight, body fat %, muscle, water, BMI, and the rest), and keeps them as an append-only log you control. It also ships an MCP server so Claude Code (or any MCP-capable agent) can answer questions about your history — "what's my weight trend this year?", "lowest body fat ever?" — by querying that log.

Demo

Exploring the data from Claude Code via the MCP server:

▶ Watch the full video for higher quality.

Related MCP server: Ultrahuman MCP Server

How it works

1. You provide your Renpho account credentials in a local .env.

2. A sync authenticates with Renpho and pulls your full reading history — the API has no date filter, so every sync fetches everything.

3. Readings are deduplicated by timestamp and appended to a raw JSONL log (the source of truth); a spreadsheet-friendly CSV is regenerated alongside it.

4. You explore the log directly (it's plain files) or via the MCP server, which exposes query/trend tools to an AI agent. The server does the math and returns data; the agent reasons over it.

Nothing leaves your machine except the calls to Renpho's own servers. There is no LLM inside this project — the "intelligence" is whatever agent you connect to the MCP server.

Quick start

pip install -e ".[dev]"          # install
cp .env.example .env             # then add your Renpho email + password
renpho-sync                      # pull your history into data/renpho_log.jsonl

Then register the MCP server (see Usage) to explore it from an AI agent.

Usage

Sync your data

Run a sync whenever you want to capture new weigh-ins:

renpho-sync                      # one-shot, no server (also: python -m renpho_health.cli)

It's safe to run repeatedly — readings are deduplicated by timestamp, so re-runs never create duplicates:

{ "pulled": 68, "new": 2, "skipped": 66, "total": 68 }

Two files are written (paths configurable — see Configuration):

• data/renpho_log.jsonl — the raw, append-only source of truth, one reading per line with every field the scale returned.

• data/renpho_log.csv — a spreadsheet-friendly view, regenerated from the JSONL on every sync.

There's also an HTTP trigger if you prefer a running service — POST /sync (plus GET /health):

uvicorn renpho_health.main:app --reload
curl -X POST http://127.0.0.1:8000/sync

Explore your data with an AI agent (MCP)

The MCP server lets an AI agent query your readings. It reads the last-synced log offline; the refresh tool pulls new readings on demand.

Register it in Claude Code (requires the claude CLI — npm install -g @anthropic-ai/claude-code). This registers it at user scope so it's available in every session, with absolute paths baked in so it works from any folder:

claude mcp add -s user renpho \
  -e RENPHO_LOG_PATH=/absolute/path/to/data/renpho_log.jsonl \
  -e RENPHO_CSV_PATH=/absolute/path/to/data/renpho_log.csv \
  -- /absolute/path/to/.venv/Scripts/renpho-mcp

Use absolute paths (the launcher is renpho-mcp.exe on Windows). Equivalently, the launch command can be python -m renpho_health.mcp_server using the venv's Python.

Start a new session, confirm with /mcp (or claude mcp list), then ask e.g. "using renpho, what's my current weight and how far am I down this year?" The server exposes these tools plus a renpho://readings resource:

Notes:

• Dates are ISO 8601 (YYYY-MM-DD) only, so there's no DD-MM/MM-DD ambiguity — the agent resolves "last 3 months" to a real date before calling.

• Units are reported per metric from a curated map (values are stored canonically, e.g. weight in kg); metrics whose unit isn't certain report null rather than guess.

• The read-only tools work from anywhere; refresh additionally needs RENPHO_EMAIL/RENPHO_PASSWORD — either run from the project folder (so .env loads) or add them as -e vars when registering. See docs/adr/0004-mcp-server.md.

Configuration

Copy .env.example to .env and fill in your Renpho account:

Your credentials and the data/ directory (personal health data) are gitignored and never leave your machine. Renpho has no official API; this talks to Renpho's app backend using a reverse-engineered protocol (see docs/adr/0002-own-renpho-client.md).

Architecture

A pull pipeline feeds a local log that two front ends read from: an HTTP endpoint / CLI for syncing, and an MCP server for querying. All the analytics live in a pure, testable module; the MCP server and HTTP app are thin wrappers over it.

src/renpho_health/
  renpho/        # in-repo Renpho API client (AES-128-ECB envelope)
  store.py       # append-only JSONL log + derived CSV, dedup by timeStamp
  sync.py        # pull -> dedup -> log pipeline (run_sync)
  cli.py         # `renpho-sync` command
  main.py        # FastAPI app: POST /sync, GET /health
  queries.py     # pure analytics over the log (stats, trend, history)
  mcp_server.py  # MCP server: tools + renpho://readings resource
tests/           # pytest suite (no network — fakes the Renpho client)
docs/            # specs, ADRs, devlog

Roadmap

• Sync the log into a destination of your choosing — a Google Doc or a Markdown file in a Google Drive vault (spec).

• An automatic scheduler so syncs run without manual triggering.

Changelog & development notes

See CHANGELOG.md for release history, docs/DEVLOG.md for development history, docs/adr/ for design decisions, and docs/specs/ for feature specs.

License

MIT — see LICENSE.