@AlexP, I have to agree with @Dunky in that I thought the Outbound Site-to-Site VPN firewall was a stateful firewall. After reading @Dunky‘s post I thought I’d try it in my lab…
So I have Pi-A on 172.17.10.10/24 on site A, and another Pi, Pi-B on 172.18.10.10/24 on Site B. The two are connected via a hub and spoke Auto VPN, the hub being site H.
With no Site-to-Site VPN firewall rules I can successfully ping from 172.17.10.10 to 172.18.10.10, and also from 172.18.10.10 to 172.17.10.10 (the reverse).
I then put in place a deny rule on the Site-to-Site VPN firewall, Deny ICMPv4 from source 172.17.10.10 to destination 172.18.10.10.
I then ran the ping test again. The ping from 172.17.10.10 to 172.18.10.10 failed, as expected. However, the ping from 172.18.10.10 to 172.17.10.10 was successful, which makes me believe the Site-to-Site VPN firewall is stateful. As the first ICMP message to 172.17.10.10 was seen in the inbound direction it stored the state and allowed the return (outbound) message from 172.17.10.10 toward 172.18.10.10, even though the firewall rule should deny it if it was stateless. Therefore I believe the Site-to-Site VPN firewall is stateful.
It would be great to get confirmation of this, or if there are any flaws in my testing (other than it just being for ICMP).
@Dunky, if as per above, the Site-to-Site VPN firewall is stateful then there is no solution to your problem as the inbound connection will always establish state, and thus be permitted in the outbound direction. As @ww stated, you may be able to achieve something with a Group Policy applied to the destination client (the ACLs in them are stateless), or alternatively put a third party firewall into the solution that can apply inbound firewall rules to the VPN, which will terminate the non-Meraki VPN peer (and may also provide for other flexibility for the non-Meraki peer too).