Your I²C sensor isn’t showing up? Here’s the fix, in order

If you want the physical picture of what I²C actually does on the wire — START, ACK, clock stretching — see the interactive one-byte-six-buses explainer. This post is the practical counterpart: the bus is silent, now what?

Step 1 — Always scan the bus first

Never debug blind. Every I²C device answers on a 7-bit address, and you can ask the bus who’s home. Flash this and open the Serial Monitor at 115200:

#include <Wire.h>

void setup() {
  Serial.begin(115200);
  Wire.begin();               // ESP8266: D2/D1  ·  ESP32: 21/22  ·  Uno: A4/A5
}

void loop() {
  byte found = 0;
  Serial.println("Scanning I2C bus...");
  for (byte addr = 1; addr < 127; addr++) {
    Wire.beginTransmission(addr);
    if (Wire.endTransmission() == 0) {          // 0 = the device ACKed
      Serial.printf("  found device at 0x%02X\n", addr);
      found++;
    }
  }
  Serial.println(found ? "done." : "No devices found — check wiring & pull-ups.");
  delay(3000);
}

Three outcomes, three diagnoses:

• Your sensor’s address shows up (e.g. 0x76 for a BME280) — wiring and power are fine; the bug is in your driver code or the wrong library address.
• A different address shows up — you’ve got the wrong part or the wrong address (many boards let you move between two addresses with a solder jumper).
• “No devices found” — the electrical layer is broken. Read on.

Prefer a scanner on a web page instead of a serial cable? The RisalDash I2C-Scanner example lists every address with a best-guess of the chip, live in the browser.

Step 2 — Pull-up resistors (the #1 cause of a silent bus)

I²C outputs are open-drain: a device can only pull a line down to 0. What pulls it back up to logic high is a pair of pull-up resistors — one on SDA, one on SCL, to 3V3. Without them the lines float, nothing reads high, and the scan finds nothing at all.

• Many breakout boards already have pull-ups (often 10 kΩ or 4.7 kΩ). One sensor on a short wire usually just works.
• Add your own only if the bus master board has none. Typical values: 4.7 kΩ at 100 kHz, 2.2 kΩ at 400 kHz — two resistors for the whole bus, not per device.
• Stacked too many modules that each carry pull-ups? The parallel resistance drops too low and the lines can’t be pulled up cleanly — remove the extras.

The value is a trade-off: smaller resistance = crisper edges but more current; the datasheet limit is the bus capacitance (≤400 pF), which brings us to the next trap.

Step 3 — Long wires kill I²C

I²C was designed for chips on one board — centimetres, not metres. Every extra centimetre of wire adds capacitance, the rising edge rounds off, and past ~400 pF the master can’t tell a 1 from a 0. Symptoms: it works on the bench, then flakes out once you route the sensor across the room.

• Drop the speed to 100 kHz (or lower): Wire.setClock(100000);
• Use stronger pull-ups (2.2 kΩ) to fight the capacitance.
• Use twisted or shielded pairs, keep runs short, and don’t share a ribbon cable with noisy signals.
• Genuinely need metres? I²C is the wrong bus — use a differential one (RS-485/CAN). See why I²C is short-range.

Step 4 — Two sensors, one address

The scan shows a device, but your second identical sensor is invisible — because they share the same fixed address. Two BME280s both answer on 0x76 and clobber each other. Fixes:

• Address pin — most sensors expose an ADDR/SDO pin that flips them to a second address (e.g. 0x76 → 0x77). Tie one high, one low.
• An I²C multiplexer — a TCA9548A gives you 8 isolated sub-buses, so eight identical sensors coexist behind one address.
• A second bus — ESP32 has two hardware I²C peripherals; run Wire1.begin(sda2, scl2) on different pins.

Still nothing? The usual last suspects

• SDA/SCL swapped — the classic. SDA→SDA, SCL→SCL; crossing them looks identical but the bus stays silent.
• Voltage levels — a 5 V sensor on a 3.3 V ESP32 (or vice-versa) needs a level shifter; don’t wire 5 V straight into an ESP32 pin.
• Power — a hungry module browning out the 3V3 rail drops off the bus. Give it a solid supply and a decoupling cap.
• A stuck slave holding SDA low locks the whole bus. Recover by pulsing SCL up to 9 times before Wire.begin(), or just power-cycle.

Work the list top to bottom — scan, pull-ups, wire length, addresses, then the stragglers — and a dead I²C bus almost always comes back to life within a few minutes.