Renpho ES-CS20M BLE

This package provides an unofficial interface for interacting with Renpho's ES-CS20M scale (and other scales that share the same QN-series protocol, including some non-Renpho ones) over Bluetooth Low Energy. It also has experimental, weight-only support for a broadcast-only ES-CS20M subvariant that speaks a different protocol. See the Device compatibility section for the current list of confirmed-working models.

Disclaimer: This is an unofficial, community-developed library. It is not affiliated with, endorsed by, or connected to Renpho, its parent companies, subsidiaries, or affiliates. The official Renpho website can be found at https://www.renpho.com. "Renpho", "ES-CS20M", and other model names referenced here, along with related marks, emblems, and images, are property of their respective owners. Use of any trade name or trademark is for identification and reference purposes only and does not imply any association with the trademark holder.

Features

• Live weight and body fat readings from the scale's notification stream.
• Guest-mode protocol — coexists safely with users registered by the official Renpho app on the same scale.
• Three modes: fixed-user (with Profile), user-detection (with async resolver), and weight-only.
• BodyMetrics derives 9 body-composition metrics from a stable reading: BMI, fat-free mass, body water %, skeletal muscle %, muscle mass, bone mass, protein %, BMR, and a body fat % passthrough.
• On models that compute a full body-composition panel on-device (e.g. the R-MSB01), the scale's own metrics — BMI, body water, muscle and bone mass, visceral fat, BMR, protein and more — are delivered merged into the measurement callback.
• Experimental: weight-only support for the broadcast-only ES-CS20M subvariant via RenphoAABBScale (no body composition — see Device compatibility).

Installation

pip install renpho-escs20m

PyPI uses the hyphenated name renpho-escs20m; the import name uses underscores: import renpho_escs20m.

Device compatibility

The library speaks up to three Renpho BLE protocols; what a scale supports depends on which one its hardware uses:

Protocol support at a glance

Identifying your scale

Which protocol a scale speaks doesn't track the marketed model name: several Renpho models share the QN-series hardware, while some ES-CS20M hardware revisions speak a different (broadcast-only or not-yet-supported) protocol. The reliable discriminator is the HVIN (Hardware Version Identification Number) printed on the regulatory sticker on the back of the scale, including its trailing revision code (e.g. …MA2 vs …MB2 vs …MN). Some stickers don't print HVIN as a separate field — in that case the same identifier is usually embedded as the trailing portion of the FCC ID (e.g. FCC ID 2A26P-ESCS20MA2 → device code ESCS20MA2). The tables below list both, so you can match on either.

Confirmed-working (QN-series):

• R-MSB01 — a later hardware revision of the same CS20 platform. Extended-flavor (on-device body fat via a BLE-pushed profile), FFF0 GATT layout, identical handshake and field offsets to ES-CS20M — the one difference is its extended measurement frame is 15 bytes, not 14, with one extra byte before the checksum that the parser does not read. It also computes the full body-composition panel on-device and sends it in two further frames after each measurement, which the library reports via the on-device metrics keys described under Measurements. Its resistance fields seem to be obfuscated on the wire, so they are omitted from the measurement callback rather than reported as ohms — body composition comes from the scale itself here, so nothing is lost.

Experimental:

• Arboleaf CS20M — QN-series hardware ships the same wire protocol on two GATT service layouts, and the library handles both: the FFF0 layout used by the Renpho models above, and the FFE0 layout seen on some other QN scales, like this one. Full feature set via RenphoQNScale.
• ES-CS20M (FCC ID 2APXUES-CS20M) — a non-connectable subvariant: it broadcasts weight in its BLE advertisements rather than connecting over GATT, using a different (0xaabb) protocol. Supported via RenphoAABBScale. Impedance is not reported (for body composition calculations see Broadcast variant). The display unit can be observed but not set.

Known-incompatible — 0x55aa (not yet supported):

The Protocol column records the first bytes of the notification frames each unsupported variant emits — a rough fingerprint of the (different) BLE protocol it speaks, kept for reference and possible future support work.

The pattern so far: marketed model name is unreliable, but the HVIN — and specifically its revision suffix (A2, B2, N…) — tracks the actual hardware and apparently also the protocol. If your Renpho scale HVIN ends in A2 or N, this library will likely work with it; if it ends in some other suffix, try it out to see if it works and report back on the issue tracker.

This library may also work with other QN-Scale varieties utilizing the same protocol (on either GATT layout), including non-Renpho ones. Feel free to report compatibility results on the issue tracker.

Reporting a compatibility result

If your scale isn't in any of the tables above, open an issue at github.com/ronnnnnnnnnnnnn/renpho-escs20m/issues with:

• Marketed model (e.g., ES-CS20M)
• HVIN from the back-of-device sticker (including the revision suffix)
• Whether the library actually drives the scale correctly (live weight notifications, body fat values, etc.)

The library itself doesn't gate or warn on compatibility at runtime — it'll attempt the handshake against any device. This section is the canonical compatibility record.

Quick start

Weight only (no body fat)

import asyncio
from renpho_escs20m import RenphoQNScale, ScaleData, WEIGHT_KEY, WeightUnit


def notification_callback(data: ScaleData):
    print(f"weight={data.measurements[WEIGHT_KEY]} kg")


async def main():
    scale = RenphoQNScale(
        'XX:XX:XX:XX:XX:XX', notification_callback, WeightUnit.KG,
    )
    await scale.async_start()
    await asyncio.sleep(30)
    await scale.async_stop()


asyncio.run(main())

Fixed user + body metrics

import asyncio
from renpho_escs20m import (
    BODY_FAT_KEY, BodyMetrics, Profile, RenphoQNScale,
    ScaleData, Sex, WEIGHT_KEY, WeightUnit,
)


PROFILE = Profile(
    sex=Sex.Male,
    age=35,
    height_m=1.80,
    athlete=False,
    algorithm=0x04,        # see "Body fat algorithm" below
)


def notification_callback(data: ScaleData):
    weight = data.measurements.get(WEIGHT_KEY)
    body_fat = data.measurements.get(BODY_FAT_KEY)
    if weight is not None and body_fat is not None:
        m = BodyMetrics(
            weight_kg=weight,
            height_m=PROFILE.height_m,
            age=PROFILE.age,
            sex=PROFILE.sex,
            body_fat_percentage=body_fat,
        )
        print(
            f"weight={weight} kg  bmi={m.body_mass_index}  "
            f"bf%={m.body_fat_percentage}  bmr={m.basal_metabolic_rate}"
        )
    elif weight is not None:
        print(f"weight={weight} kg  bmi={round(weight / PROFILE.height_m**2, 1)}")


async def main():
    scale = RenphoQNScale(
        'XX:XX:XX:XX:XX:XX',
        notification_callback,
        WeightUnit.KG,
        profile=PROFILE,
    )
    await scale.async_start()
    await asyncio.sleep(30)
    await scale.async_stop()


asyncio.run(main())

User detection from weight

import asyncio
from renpho_escs20m import (
    Profile, RenphoQNScale, ScaleData, Sex, WEIGHT_KEY, WeightUnit,
)


KNOWN_USERS: dict[str, Profile] = {
    'alice': Profile(sex=Sex.Female, age=34, height_m=1.65),
    'bob':   Profile(sex=Sex.Male,   age=43, height_m=1.78),
}


async def resolve_user(weight_kg: float) -> Profile | None:
    """Pick the user whose typical weight is closest to the reading.

    Real implementations would do a DB lookup, talk to a Home
    Assistant entity, etc. The callback is async so I/O won't block
    the BLE event loop.
    """
    if weight_kg < 70:
        return KNOWN_USERS['alice']
    return KNOWN_USERS['bob']


def notification_callback(data: ScaleData):
    print(f"weight={data.measurements[WEIGHT_KEY]} kg")


async def main():
    scale = RenphoQNScale(
        'XX:XX:XX:XX:XX:XX',
        notification_callback,
        WeightUnit.KG,
        profile=resolve_user,        # ← user-detection mode
    )
    await scale.async_start()
    await asyncio.sleep(30)
    await scale.async_stop()


asyncio.run(main())

The scale firmware will not start a measurement without a profile reply, so the library always sends one in response to the scale's 0x21 05 ff profile request. In detection mode it sends a bootstrap profile with algorithm=0x00 (body fat calculation disabled) so the measurement starts; on the first stable weight frame it awaits resolve_user(weight) and writes the returned profile to the scale, which then computes body fat and emits the stable-with-metrics frame. Returning None from the resolver leaves the bootstrap profile in place — the scale stays in weight-only mode for that session.

The resolver must return faster than the scale's internal body fat commit window — empirically about 2 seconds after the first stable frame. If it doesn't, the scale will finalize the measurement against the bootstrap profile (no body fat) before your resolved profile lands. If the BLE session ends while the resolver is still in flight, the library cancels the resolver task to avoid leaking work.

Broadcast variant

The broadcast-only 0xaabb subvariant uses a different client, RenphoAABBScale — no Profile, no unit control, weight only:

import asyncio
from renpho_escs20m import RenphoAABBScale, ScaleData, WEIGHT_KEY


def notification_callback(data: ScaleData):
    print(
        f"weight={data.measurements[WEIGHT_KEY]} kg  "
        f"(scale display shows {data.display_unit.name})"
    )


async def main():
    scale = RenphoAABBScale('XX:XX:XX:XX:XX:XX', notification_callback)
    await scale.async_start()
    await asyncio.sleep(30)
    await scale.async_stop()


asyncio.run(main())

Estimating body composition. This scale transmits no impedance, so RenphoAABBScale reports weight only. You can still produce body-composition metrics by calling calculate_body_fat() / BodyMetrics with a fixed synthetic impedance (~500 Ω): the algorithm is near-impedance-independent in that band — verified <1 percentage-point change in body fat across R = 300–900 Ω — so the exact value barely matters, and ~500 reproduces pretty closely what the official Renpho app shows.

Protocol detection

detect_protocol() classifies a BLE advertisement (local name, manufacturer data, and address) as ScaleProtocol.QN, ScaleProtocol.AABB, or None if it isn't a recognized scale. Pair it with SCALE_CLASSES to pick the right client class without hardcoding if/else branches:

from renpho_escs20m import SCALE_CLASSES, detect_protocol

protocol = detect_protocol(local_name, manufacturer_data, address)
if protocol is not None:
    scale_cls = SCALE_CLASSES[protocol]
    scale = scale_cls(address, notification_callback, ...)

Frame layouts (manufacturer-data value, company ID already stripped):

• QN: [0:2] model identifier, 16-bit big-endian; [2:5] varies per advertisement; [5:11] device MAC address, little-endian.
• AABB: [0:2] 0xAABB magic; [2:8] device MAC address, forward byte order; [8:] protocol payload.

Known QN model identifiers, observed in real advertisement captures:

Advertisements with an unrecognized QN model identifier still classify via a name/address fallback matcher and log a warning — reporting that warning (or an incompatible scale on the issue tracker) is how the identifier registry grows.

API reference

Scale client

• RenphoQNScale(address, callback, display_unit, *, profile=None, scanning_mode=BluetoothScanningMode.ACTIVE, …) — BLE scale client. The profile argument is one of:

  • a Profile (fixed-user mode),
  • a ProfileResolver (user-detection mode),
  • None (weight-only mode, default).

  The profile modes only apply to the QN extended flavor (e.g. HVIN ESCS20MA2) — the variant where the scale computes body fat on-device from a profile sent over BLE. The basic flavor (HVIN ESCS20MN) takes no profile over BLE, so the argument is ignored there: readings always include weight plus raw impedance, and body fat is computed off-scale via calculate_body_fat(). In all cases "weight-only mode" refers to what the scale computes and displays — it never restricts what the library reports: both flavors deliver raw impedance in every mode, and it is passed through — except on models whose resistance fields are obfuscated on the wire, where the values are withheld (see Measurements).

  clear_stored_measurements=True (default False) drains the scale's store of offline measurements — readings taken while nothing was connected — once per session. Receiving a stored reading deletes it from the scale (the protocol has no separate delete command), so enabling this hides those readings from any other client: leave it off if you also sync the scale with the official Renpho app. Drained readings are logged at debug level and discarded for now. Each flavor is queried with its own command form.

  Additional keyword arguments (adapter, cooldown_seconds, max_connect_attempts, bleak_scanner_backend, logger) are available for advanced use — see the class docstring. RenphoESCS20MScale remains importable as a backward-compatible alias for RenphoQNScale.

• callback (passed to RenphoQNScale) — invoked once per measurement, on the final frame the scale emits (the stable-with-metrics frame on the extended flavor; the status-0x01 final frame on the basic flavor). On recognized panel-sending models (e.g. the R-MSB01) the invocation comes a moment later, once the scale's body-composition panel has arrived and been merged in — see Measurements. In user-detection mode, the earlier stable frame is used only to trigger the profile resolver and does not reach the callback. Within the frame, ScaleData.measurements always contains WEIGHT_KEY; BODY_FAT_KEY and the two RESISTANCE_*_KEY entries are present only when the scale actually produced non-zero values for them. BODY_FAT_KEY requires the extended flavor and a real profile: it is absent in weight-only mode, in user-detection mode if the resolver returned None, and any time algorithm=0x00. Impedance is reported by both flavors in every mode — the impedance pass runs even under the bootstrap (algorithm=0x00) profile — but withheld on obfuscated-resistance models. The basic flavor never produces BODY_FAT_KEY — compute body fat from RESISTANCE_1_KEY with calculate_body_fat().

• scale.battery_level — last successfully-read battery percentage (int | None). May be None until first successful read. Reliability caveat: on at least one observed unit (firmware V10.0) the scale reported a static 100 and didn't appear to decrement it as the batteries drained — reading 100% even on cells weak enough to need replacing — and exposed no other battery source over BLE. It's unknown whether other hardware revisions or firmware behave the same way, so treat a steady 100% as possibly unreliable rather than assuming it; the value is reported as-is and may be accurate on your device.

• scale.firmware_revision — last successfully-read firmware revision string (str | None). May be None until first successful read or when response is empty.

• BluetoothScanningMode — ACTIVE (default) / PASSIVE, passed via the scanning_mode kwarg. PASSIVE only takes effect on Linux (BlueZ); other platforms fall back to active.

Broadcast variant (experimental)

• RenphoAABBScale(address, callback, *, scanning_mode=…, adapter=…, bleak_scanner_backend=…, logger=…) — client for the non-connectable 0xaabb ES-CS20M subvariant. It never opens a GATT connection; it reads weight straight from the scale's BLE advertisements. Differences from RenphoQNScale:
  • ScaleData.measurements contains only WEIGHT_KEY (always kg).
  • ScaleData.display_unit reflects the unit the scale's LCD is showing (observed from the advertisement). It is read-only — the scale cannot be told to change units, and assigning display_unit is ignored.
  • No profile, no body composition (this scale does no impedance/BIA), and no battery_level / firmware_revision.
  • Weight-only, and validated against captured advertisements rather than live hardware.

Extending the library

• RenphoScale / GattScale / AdvertisementScale — the abstract base classes the concrete clients subclass (RenphoScale holds the scanner lifecycle; GattScale and AdvertisementScale are the connection-based and advertisement-based transports). Exported for adding new protocol variants.

Profiles

• Profile(sex, age, height_m, athlete=False, algorithm=0x04) — user-profile inputs the scale needs to compute body fat on-device (extended flavor; the basic flavor takes no profile over BLE — feed the same inputs to calculate_body_fat() instead). See Profile's docstring for the wire semantics of each field.
• ProfileResolver — type alias for the async callback used in user-detection mode: Callable[[float], Awaitable[Profile | None]]. Receives the first stable weight in kg and returns the Profile to write (or None to skip). Extended flavor only — the basic flavor never requests a profile, so the resolver is never invoked for it.

Measurements

• ScaleData — dataclass passed to the notification callback. Fields: name, address, display_unit, and measurements (a dict keyed by the constants below).

• WeightUnit — KG, LB, ST, ST_LB.

• Measurement-dict keys (constants importable from renpho_escs20m):

  • WEIGHT_KEY ("weight") — kg
  • BODY_FAT_KEY ("body_fat") — % (extended flavor only, and only when the scale ran its on-device body fat calculation)
  • RESISTANCE_1_KEY, RESISTANCE_2_KEY ("resistance_1", "resistance_2") — bioelectrical impedance in ohms (present on final frames when non-zero; the two readings are typically within a couple of ohms of each other and either can be fed to calculate_body_fat()). Absent on models that obfuscate these fields, where the wire values are not ohms.

• On-device body composition, reported only by models that compute it themselves (see below):

  • BMI_KEY, BODY_SCORE_KEY ("bmi", "body_score") — ratings
  • BODY_WATER_KEY, PROTEIN_KEY, SUBCUTANEOUS_FAT_KEY, SKELETAL_MUSCLE_KEY — %
  • MUSCLE_MASS_KEY, BONE_MASS_KEY, FAT_FREE_MASS_KEY — kg
  • BMR_KEY — kcal/day
  • VISCERAL_FAT_KEY, BODY_AGE_KEY, BODY_SHAPE_KEY — the scale's own unitless ratings. BODY_AGE_KEY is the scale's value; the Renpho app displays a metabolic age it derives separately, so the two differ slightly.

  The scale sends its body-composition panel a moment after the final measurement frame. On recognized panel-sending models the library holds the reading briefly and delivers it once, complete — one weigh-in, one callback, panel included. If the panel never finishes (lost frames, early disconnect), the reading is still delivered within about a second, just without the missing fields — it is never lost. A model that sends these frames without being recognized gets its reading immediately as usual, minus the panel: the frame layout is only trusted on verified models, so the library logs a warning asking for the model to be reported rather than guessing.

Body composition

• BodyMetrics(weight_kg, height_m, age, sex, body_fat_percentage) — derives body-composition metrics from a stable reading. Call it from the notification callback once a Profile is known. No athlete parameter: by the time a body fat value reaches this class, the scale's firmware has already applied the athlete adjustment. Exposes these snake_case attributes:
  • body_mass_index — BMI
  • body_fat_percentage — passthrough of the constructor input
  • fat_free_mass (kg)
  • body_water_percentage
  • skeletal_muscle_percentage
  • bone_mass (kg)
  • muscle_mass (kg)
  • protein_percentage
  • basal_metabolic_rate (kcal/day, integer)
• calculate_body_fat(weight_kg, height_m, age, sex, resistance, *, algorithm=0x04, athlete=False) — off-scale approximation of the on-device body fat formulas (algorithms 0x03 and 0x04 only). Complements BodyMetrics: BodyMetrics takes an already-computed body fat value as input, while calculate_body_fat computes one from raw impedance. The typical pairing is to feed calculate_body_fat's output into BodyMetrics when a slow user-detection lookup misses the scale's commit window and body fat needs to be recomputed from RESISTANCE_1_KEY after the fact.

Low-level

• build_user_profile_command(...) — raw command builder for the guest-mode user-profile frame the scale expects. Most callers should construct a Profile and let RenphoQNScale call this builder; use it directly only if you need to bypass the protocol state machine.

Body fat algorithm (Profile.algorithm)

Selects which on-device body fat formula the scale runs. Most callers should leave this at the default.

• algorithm=0x04 (default) and algorithm=0x03 are the two formulas Renpho's app selects from in normal use. The selection appears to depend on user region.
• algorithm=0x00 disables the on-scale body fat calculation entirely — the measurement itself still runs; the scale just computes and displays no body fat. This is what the library uses internally in weight-only mode and during user-detection bootstrap.
• Other values (0x01, 0x02, 0x05, 0x06) are accepted by the scale but don't seem to be used by Renpho's app and aren't validated against it — treat them as experimental.

Profile.athlete=True is independent of algorithm: it switches the firmware to its athlete-tuned curve regardless of which formula is selected.

The library also ships an off-scale approximation of algorithms 0x03 and 0x04 via calculate_body_fat() — useful when the scale's body fat commit window closes before a slow user-detection lookup resolves. The other algorithms aren't currently approximated in software.

App-matching conventions

The Renpho app applies a few non-obvious transformations to profile data before running the body fat calculation. The library diverges from one and leaves the other to the caller:

1. Height precision: library passes through; app truncates to whole cm. The Renpho app truncates a 170.7 cm profile to 170 cm before running the body fat calculation. This library passes the user's exact height_m through to the scale (rounded to the nearest mm), giving slightly more precise body fat from the scale's on-device curve.
   • If you want to reproduce the Renpho app's displayed values exactly (for cross-checking), pre-truncate the call site: height_m = int(actual_cm) / 100.
2. Age is birthday-aware. For a profile whose UI age shows N, the app uses N if the birthday has already occurred this year, else N − 1. Profile.age is a plain integer — callers wanting to match the app should compute this themselves before constructing the Profile.

Platform compatibility

• Python 3.11+
• bleak 2.x or 3.x (bleak>=2.0.0,<4.0.0)
• Tested on macOS (Apple Silicon)
• Linux via BlueZ should work through the standard bleak backend but is unverified
• Compatibility with Windows is unknown

Troubleshooting

On Raspberry Pi (and possibly other Linux machines using BlueZ), if you encounter a org.bluez.Error.InProgress error, try the following in bluetoothctl:

power off
power on
scan on

(See home-assistant/core#76186 (comment) for context.)

Acknowledgments

• R-MSB01 support contributed by @Jaano — thank you!

Support the project

If you find this unofficial project helpful, consider buying me a coffee! Your support helps maintain and improve this library.

License

This project is licensed under the MIT License - see the LICENSE file for details.