The chains use the CMB data from WMAP. The CosmoMC paper appendices describe importance sampling in detail, which is what allows you to combine other constraints with WMAP using these chains.
Example
Let's say you have a new constraint on sigma8, sigma8= 0.7+-0.05
(1sig, Gaussian), and
you want to see what the posterior of the parameters is using your
constraint, and WMAP. All you need to do is read in each line of
the WMAP samples text file,
multiply the weight of each sample by
exp(-(sigma8-0.7)**2/(2*0.05**2)), the correction to the posterior for each sample's
different sigma8. The new mean of n_s, for example, would then be
sum(n_s*weight)/sum(weight).
The better your constraint is the noisier the output will be, and if
you find that only a very small fraction of the samples get
significant weight then the results cannot be trusted and you will
have to MCMC from scratch.
CosmoMC comes with a handy program,
"getdist" for processing your resultant text file and computing
statistics, and plots.
The text files are supplied as produced by cosmomc. See the CosmoMC readme for details. The normalization of the weights is arbitary, as is the origin of the ln(likelihood). Om_k=0, w=-1, A_t=0, f_v=0 are assumed, plus all the other standard assumptions.
Download here (WMAP only, flat prior on H_0 40-100).
If you use these chains you should cite the WMAP papers by Verde, Hinshaw and Kogut, and the B03 paper (which is what these chains were made for).