Raspberry Pi Soil Moisture Sensor Setup for Automated Garden Monitoring with OpenClaw

Resistive soil moisture probes fail within 4 to 12 weeks of continuous use in moist soil. A 2024 IEEE study on sensor power supply methods confirmed that DC current flowing between exposed electrodes causes "electrochemical processes at the electrode-soil interface," building up oxides that destabilize contact resistance until readings become meaningless. Adding fertilizer accelerates the timeline because dissolved ions increase conductivity and speed up electrolysis.

SwitchDoc Labs documented the same problem in their Raspberry Pi sensor comparison: resistive probes "suffered electroplating damage within months" and produced "weird bad data" when running more than one sensor simultaneously. Capacitive sensors, by contrast, showed no corrosion and returned consistent readings across all channels. The difference is structural. A capacitive sensor measures the dielectric permittivity of the surrounding soil through a sealed PCB. Water has a dielectric constant around 80, dry soil sits near 4, and the sensor's onboard circuit outputs an analog voltage proportional to that change. No current passes through the soil, so there is nothing to corrode.

The tradeoff is that capacitive sensors output an analog voltage, and the Raspberry Pi has no built-in analog-to-digital converter. You need an external ADC chip like the MCP3008 or ADS1115 to translate that voltage into a number your Python script can read. Most beginner tutorials either skip this detail or gloss over calibration, which means the readings they produce are raw numbers disconnected from actual soil moisture. This guide covers the full stack: wiring, reading, calibrating, and then adding an OpenClaw agent that automates the monitoring and alerting that a bare Python script cannot sustain.

Before writing any code, gather four components.

Parts list

• Raspberry Pi with GPIO headers (Pi 3B+ through Pi 5 all work)
• Capacitive soil moisture sensor v2.0 (typically $3 to $5)
• MCP3008 ADC chip (10-bit, 8-channel, SPI) or ADS1115 (16-bit, 4-channel, I2C)
• Breadboard and jumper wires
• 5V relay module and 12V submersible pump (optional, for automated watering)

Choosing an ADC: MCP3008 vs ADS1115

The MCP3008 communicates over SPI, gives you 8 analog channels at 10-bit resolution (values from 0 to 1023), and costs about $2. It is the most common choice in Raspberry Pi soil sensor projects because the spidev Python library ships with most Pi OS images.

The ADS1115 uses I2C, provides 4 channels at 16-bit resolution (values from 0 to 65535), and includes a programmable gain amplifier that lets you read small voltage differences more accurately. It costs about $4. AB Electronics sells a Pi-compatible ADC board based on this architecture that plugs directly into the GPIO header. If you only need one or two sensor channels and want higher precision, the ADS1115 is the better pick.

MCP3008 wiring to the Raspberry Pi

Connect the MCP3008 to your Pi's SPI bus:

• VDD and VREF to 3.3V (Pin 1)
• AGND and DGND to GND (Pin 6)
• CLK to SCLK (Pin 23, GPIO 11)
• DOUT to MISO (Pin 21, GPIO 9)
• DIN to MOSI (Pin 19, GPIO 10)
• CS/SHDN to CE0 (Pin 24, GPIO 8)

Then connect the capacitive soil moisture sensor:

• VCC to 3.3V
• GND to GND
• AOUT to CH0 on the MCP3008

The sensor draws about 5mA, so the Pi's 3.3V rail can power several sensors simultaneously without a separate supply.

Enable SPI on your Pi first. Open a terminal and run:

sudo raspi-config

Navigate to Interface Options, then SPI, and select Yes. Reboot if prompted.

Install the spidev library if it is not already present:

pip install spidev

Here is a minimal script that reads the capacitive sensor on MCP3008 channel 0 and prints the raw value and voltage every second:

import spidev
import time

spi = spidev.SpiDev()
spi.open(0, 0)
spi.max_speed_hz = 1350000

def read_channel(channel):
    adc = spi.xfer2([1, (8 + channel) << 4, 0])
    data = ((adc[1] & 3) << 8) + adc[2]
    return data

while True:
    raw = read_channel(0)
    voltage = raw * 3.3 / 1023
    print(f"Raw: {raw}, Voltage: {voltage:.2f}V")
    time.sleep(1)

The read_channel function sends a three-byte SPI command. The first byte is the start bit. The second byte selects which of the eight MCP3008 channels to read. The response comes back in bytes two and three: a 10-bit value from 0 (0V) to 1023 (3.3V).

If you chose the ADS1115, install the Adafruit CircuitPython library instead:

pip install adafruit-circuitpython-ads1x15

import board
import busio
import adafruit_ads1x15.ads1115 as ADS
from adafruit_ads1x15.analog_in import AnalogIn

i2c = busio.I2C(board.SCL, board.SDA)
ads = ADS.ADS1115(i2c)
chan = AnalogIn(ads, ADS.P0)

print(f"Raw: {chan.value}, Voltage: {chan.voltage:.3f}V")

Run the script with the sensor in dry air first, then push it into moist soil. You should see the voltage drop as moisture increases. Write down both readings. You will need them for calibration.

A capacitive sensor outputs a higher voltage when dry and a lower voltage when wet. The exact range depends on your specific sensor batch and your Pi's 3.3V rail accuracy. Two sensors from the same manufacturer can produce different absolute readings, which is why calibration matters and why a hardcoded threshold copied from a tutorial rarely works for long.

Raw ADC values tell you nothing about actual soil moisture. A reading of 450 could mean bone-dry sand or damp clay depending on your sensor and soil type. Calibration maps your sensor's output range to volumetric water content (VWC): the ratio of water volume to total soil volume, expressed as a percentage.

Measuring reference values

Take two reference readings. Hold the sensor in open air and record the voltage. This is your V_DRY baseline, typically around 2.85V for a capacitive v2.0 sensor on a 3.3V supply. Then submerge the sensor in a glass of water up to the marked line and record the voltage. This is your V_WET floor, typically around 1.35V.

SwitchDoc Labs documented raw 16-bit ADC readings of 22,795 for dry soil and 12,337 for saturated soil during their capacitive sensor testing. That 46% signal swing maps to the full moisture range only after applying a calibration formula.

The calibration formula

AB Electronics provides the standard linear conversion used in most soil sensing projects:

V_DRY = 2.85
V_WET = 1.35

def moisture_percent(voltage):
    vwc = ((V_DRY - voltage) / (V_DRY - V_WET)) * 100
    return max(0.0, min(100.0, vwc))

The function clamps output between 0% and 100%. Most garden plants thrive when soil VWC stays between 20% and 45%, depending on species and soil composition.

Soil-specific adjustments

Clay soils retain more water at the same VWC percentage than sandy soils, and organic matter shifts the dielectric response. For better accuracy, mix a known volume of water into a known volume of your actual garden soil (for example, 20mL water per 100mL soil = 20% VWC) and verify that your calibration formula agrees. Adjust V_DRY and V_WET if the calculated percentage diverges from the known value by more than a few points.

Persist your garden sensor data across sessions and devices

Fast.io gives your OpenClaw agent a cloud workspace with built-in search across your sensor logs and MCP server access. Starts with a 14-day free trial.

Start 14-Day Trial

Reading sensor data from a terminal is useful for testing but impractical for daily garden care. You need something that runs continuously, interprets trends, and contacts you when the soil is too dry or a sensor goes offline. An OpenClaw agent handles this.

OpenClaw is an open-source AI agent with over 160,000 GitHub stars that runs on your own hardware, including the same Raspberry Pi reading your soil sensor. It communicates through messaging apps like Telegram, Discord, and WhatsApp, which means you can receive a "soil moisture dropped below 25%" alert on your phone without building a custom notification system.

What the agent does. The agent reads the sensor values your Python script produces, compares them against your calibrated thresholds, and decides whether to act. When moisture falls below a set point, it can trigger a relay-controlled water pump. When moisture stays consistently high after rain, it skips watering. It can also detect sensor faults: if readings suddenly jump to 0V or 3.3V, the sensor likely disconnected or failed, and the agent alerts you instead of blindly watering.

Relay wiring for automated watering. Connect a 5V relay module to your Pi:

• VCC to 5V (Pin 2)
• GND to GND (Pin 9)
• IN to GPIO 17 (Pin 11)

Wire your pump's power supply through the relay's COM and NO (normally open) terminals. The pump gets its own external power source, not the Pi's.

import RPi.GPIO as GPIO
import time

RELAY_PIN = 17
GPIO.setmode(GPIO.BCM)
GPIO.setup(RELAY_PIN, GPIO.OUT)

def water_plants(seconds):
    GPIO.output(RELAY_PIN, GPIO.HIGH)
    time.sleep(seconds)
    GPIO.output(RELAY_PIN, GPIO.LOW)

The OpenClaw agent calls this watering function based on soil readings and time-of-day rules you configure through your messaging app. Water in the early morning when evaporation is lowest, skip watering if the forecast predicts rain, and log every decision for later review. Because the agent runs locally on the Pi, it keeps working even if your internet connection drops. Watering decisions happen on-device; only alerts and data logging depend on connectivity.

Local data logging on a Raspberry Pi has two failure modes that surface quickly. SD cards fill up: a sensor reading every 30 seconds generates about 2.5 million rows per year, and most Pi setups write to a SQLite file or CSV on the same card running the OS. You also cannot query the data remotely without setting up SSH tunnels or a web server, both of which add attack surface to a device sitting outside.

A cloud workspace solves both problems. Your OpenClaw agent can push structured moisture readings to a shared workspace where they persist independently of the Pi's SD card. If the Pi loses power or the card corrupts, the historical data survives.

Fast.io provides a workspace that agents access through a MCP server. The agent writes timestamped JSON or CSV files to a workspace folder. Once uploaded, Intelligence Mode auto-indexes the files, which means you can ask natural-language questions about your garden data: "What was the average moisture level last Tuesday?" or "Show me all readings below 20% in the past month." No separate database or vector store required.

Alternatives exist for this layer. You can log to a local InfluxDB instance and visualize with Grafana, push to an S3 bucket and query with Athena, or use Google Sheets via the Sheets API. Each adds its own infrastructure overhead. The advantage of a workspace like Fast.io is that it combines storage, search, and agent access in one layer, and every org starts with a 14-day free trial.

The ownership transfer feature is useful if you are building a garden monitoring system for someone else. Set up the entire workspace with historical data, sensor configuration notes, and threshold documentation, then hand the workspace to the new owner. You keep admin access for ongoing maintenance while they get full control of the data.