One thing I don't see explicitly mentioned is that a current transformer is an AC device. It measures only AC current, not DC. And produces an AC output voltage. Unless you've centered the output signal in the middle of your ADC range (typically 0V-5V), the negative portion will go below 0V, get clamped by either the protection diodes you've added or those within the MCU's pin protection circuitry, and be unreadable.
Here is possibly the simplest way to connect a current transformer to an ADC pin, and it is the method I'm currently using with success. I use a 3.3V PIC, but I've altered my schematic to be appropriate for a 5V AVR:
L1: Represents the secondary of the current transformer (with the primary being the wire you pass through it to be measured, not shown).
R1 & R2: A voltage divider, centering the voltage of L1 at 2.5V.
My current transformer has a turn ratio of 1,000. So for every 1A of current in the primary, it produces 1A/1000=1mA of output on the secondary. To measure that with an ADC, I need to convert that current to a voltage, and that's done simply by putting a resistor across the leads of the current transformer secondary, R3. But what value does it need to be?
The device under measurement is designed for up to 5A, and is fused at that. But it's possible it may go higher without blowing the fuse. And if it does I want to be able to detect that, without having the measurement clipped, so the MCU can shut down the device. So I allow measurement up to 10A. That reduces the resolution of the ADC by half, but that's acceptable to me.
Time for good old Ohm's Law. Since the AC output from the transformer will be superimposed on the 2.5VDC voltage from the divider (R1/R2), it can swing up to +/- 2.5V in either direction before it exceeds the MCU's 0-5V range; so E=2.5. At 10A, the transformer produces 10mA; so I=0.01. We want to know R, so we use Ohm's Law form R=E/I, resulting in 250=2.5/0.01. R3 needs to be a 250ohm resistor. That's not a standard value, and I don't need exactly 10A range; so in reality I just use the closest value I have on hand, and adjust the ADC reading-to-current conversion code accordingly.
For better accuracy, I use 1% tolerance resistors for R1, R2, and R3. And we're done. Well, sort of.
Even though I've provided a 2X safety margin, Murphy's Law says that at some point I'll get a larger surge. Maybe one that will very briefly drive current up to 20A before the fuse blows - resulting in an theoretical output of 10V (or -7.5V). In the real world it will actually be less, since my current transformer is only rated for 5A, its output drops off above that. And the MCU's built-in clamping diodes should handle it, it's precisely their job to handle brief transients. But I want just a little more reassurance.
At this point I could add two external clamping diodes, as [Jeroen3] showed in his schematic.
Instead, I added R4 (3.3K), which will reduce the peak current flowing through the MCU's clamps. At 10V, and assuming 5.7V clamping voltage, that'll be a mere (10-5.7)/3300=1.3mA through the MCU's internal clamp for a very brief period. Not as good as if I've provided external clamps, but enough to satisfy my paranoia.
How did I choose 3.3K? Somewhat arbitrarily. I was already using 3.3K for R1/R2, so it simplifies construction to reuse the same value. I did at least make sure that the value is low enough that according to my MCU's documentation, it shouldn't interfere with the charging of the ADC's S/H capacitor. In case I'm wrong about that (or perhaps your AVR requires a lower impedance), I also added C1. Being several orders of magnitude larger than the size of the ADC's S/H cap, it will stabilize the voltage even if R4 is too large, so long as samples aren't taken at the ADC's maximum rate. Assuming you're measuring a common 50/60hz signal like me, you won't need extremely high sample rates anyway.
Finally, R4/C1 form an RC filter. So in addition to providing extra protection (eliminating two external diodes), and impedance matching (eliminating an op-amp buffer), they also help to filter any high frequency noise picked up from nearby digital circuits, which improves ADC reading accuracy. Pretty good for just two extra components.
Your specific requirements (protection, frequency, accuracy, etc) may differ from mine, and I don't consider myself a guru on this topic. So take my advice accordingly. And I welcome peer review of what I've presented.