Hi everyone,

In my apartment MT AP is installed in the living room. Unfortunately space is so crowded with neighbor's WiFi networks that signal barely reaches 5 mbps in my bedroom.
Thankfully I happened to have another MT router which I can connect over wire to the first one.

In a nutshell I have two MT routers: hAP ac lite (hAP) and RB951-2n (RB). Both use RouterOS 6.36. hAP is the main AP connected to internet through ether1. RB is connected to hAP through ether1. Both routers have ether2-master and wlan under a bridge.

My end goal is some kind of a virtual router with one ethernet port connected to internet and all other interfaces under the same bridge with a seamless WiFi.

My configuration so far is the following:

hAP

- ether1 is connected to internet
- bridge of wlan1 (2.4ghz), wlan2 (5ghz) and ether2-master
- DHCP Server that serves that bridge
- CAPsMAN which has configuration of WiFi identical to wlan1
RB

- ether1 is connected to hAP
- DHCP Client configured on ether1
- bridge of wlan1 (2.4ghz), ether1 and ether2-master
- All Firewall rules are disabled
- CAP:
Questions:

1. How can I make CAPsMAN to also manage its own wlan interfaces?
2. What is the best Tx-Power for RB951-2n assuming it serves very small area (5m x 5m)?
3. Bandwidth test (hAP -> RB) shows about 30 mbit for single direction test, 25 mbit for bidirectional tests with 100% CPU load on RB. What can be wrong?
4. What can I improve in the layout?

1. How can I make CAPsMAN to also manage its own wlan interfaces?

You need to tell the local wireless interfaces (on every CAP) that they are managed by CAPsMAN:

2. What is the best Tx-Power for RB951-2n assuming it serves very small area (5m x 5m)?

The easy option is just to let CAPsMAN automatically set your frequency, channel width, and tx power.
If you don't like easy options, I'd set the antenna gain setting up around 10dBi, which will reduce tx power accounting for multiple chains and the like - though you need to set that on the interface before enabling `/interface wireless cap`

3. Bandwidth test (hAP -> RB) shows about 30 mbit for single direction test, 25 mbit for bidirectional tests with 100% CPU load on RB. What can be wrong?

I suspect you're hitting a Forwarding vs Routing problem, and likely have a bit of the observer effect.
It's always best to do bandwidth tests between devices on either side of the link you're testing. Hopefully you have 2 devices that are powerful enough for the CPU not to get in the way that you can plug into either end?

4. What can I improve in the layout?

In your CAPsMAN config, you may want to allow Local Forwarding in the Datapath, so that traffic is bridged. Otherwise all the traffic is tunnelled from the CAP to CAPsMAN. While tunnelling is good if you have a complicated network to manage and need a central point to firewall and control access before it goes out to your networks, local forwarding takes MUCH less CPU time.
I would only turn off local forwarding if I either didn't care about data throughput or had a powerful device (RB3011, RB1100, CCR, x86).
We run an x86 CAPsMAN in VMware with about 40 CAPs inside and outside tunnelling up to 150 clients moving around the site with remote desktop connections. The CAPs (Mostly RB912) tick along at around 5-15% CPU usage when in use.