ESD is tested both directions so it doesn't much matter there, and it's down to what the nominal voltage range of the node is.
You can use a clamp diode when the voltage range is above or below a supply rail.
You can use unidirectional TVS when the voltage range would permit one clamp, but another rail is unavailable for clamping positive and negative (or various other limitations).
You can use bidirectional TVS when the voltage range is symmetrical around one rail.
If the node's range is fully constrained positive and negative, then you're done. So, two clamp diodes between supply rails, one bidirectional to ground, etc.
When ranges are not symmetrical around one rail, you can use dissimilar unidirectionals in anti-series. To, uh save cost maybe I suppose, you can use diodes to clamp to rails when applicable, along with either type of TVS as suitable. Which can be regular diodes (BAT54S, BAV99, etc.), say when you have them elsewhere in the design, or they might be cheaper in some cases.
Mind that, as you make series combinations, of course the ESL increases, which is to say the peak clamping voltages add. But adds up faster because you're also adding trace length to connect them up.
This works fine for things other than ESD, of course. Sometimes you might use a TVS in series with a diode to get well-defined flyback voltage on a relay or solenoid driver (and occasionally SMPS, like monolithic offline regulators often suggest this), and the MOSFET (if it's big enough, anyway -- a few amps say) body diode handles the negative direction.
To clarify, MOSFETs handle ESD fine so long as it either causes avalanche within energy and current ratings, or body diode forward bias. That is, through the channel specifically. It's the gate that's vulnerable to voltage alone. Note that ESD through the D-S or S-D path could still cause lift of the source, reversing the gate and causing destruction. So keep source inductance low, or allow gate to rise together with it (e.g. differential drive with Kelvin connection), if this is a necessary feature.
Note that schottky aren't actually very good with ESD; they have higher ESR than PN diodes, and in fact generally integrate PN diodes (guard rings), which activate at higher voltage drop (which is mainly for avalanche capability AFAIK, but helps in forward bias as well). I don't know what BAT54S for example is actually rated at (no one provides these data). (I can say they've passed 8/15kV 10 pulses +/- plenty of times.) But they're convenient, and maybe you need the lower signal-clamping voltage, like for using logic gates (GND/VCC clamping, or GND unidirectional type) with certain circuits that produce excess voltage (high to low level conversion, RC timer, etc.).
Tim