Unless requested otherwise, when performing two-tone measurements, I explicitly state that input/output power is "per tone"; I also note the tone spacing and that the IM products are "relative to one of two equal tones X MHz apart". I typically only use total power in efficiency calculations.
Spectrum analyzers typically put their markers on the tones, not the imaginary point 3dB higher, when displaying IM data. Saves a lot of mental math and confusion.

In the '80s some co-workers of mine were investigating microwave diathermy (microwave cooking, essentially) for prostate cancer, with microwave power delivered via a 2450MHz coaxial dipole antenna inserted into the rectum. The antenna was mounted non-concentrically in a PTFE outer covering; the mounting gave a directional characteristic to the antanna to concentrate the RF power, of approximately 10 watts, towards the prostate. Apparently this technique is now done transurethrally. Eww.
One of their photographs showing a Microlab-FXR double-slug tuner (tuned for maximum power transfer) protruding from the surgical drapes was quite hilarious, although probably not so amusing to the patient.

Having said that, DO NOT connect anything to the "antenna cap", which sounds like the anode connection, which has several thousand volts on it!
The RF output from a magnetron is generally a waveguide; the technology of waveguide-to-coaxial transitions is well-understood.
Of course you will need to properly shield the food being cooked so as to avoid cooking the cook as well.