Reading their docs, it seems like this uses Trust-on-first-use key exchange, so if there's an attacker passively observing the network at that point, they can MITM all future communications.
Also, the Autocrypt Level 1 spec, which this seems to implement, appears to be based on PGP, with the following caveat: "Sometimes Autocrypt recommends to send cleartext mail even though encryption appears technically possible."
Yes, if the attacker is strictly passive, tofu with public key should be fine. That the attacker is strictly passive is a pretty strong assumption though!
Also, the Autocrypt Level 1 spec, which this seems to implement, appears to be based on PGP, with the following caveat: "Sometimes Autocrypt recommends to send cleartext mail even though encryption appears technically possible."