In observational studies of program effects, treatments are often administered at a group level such as a school or hospital but interest focuses on effects at an individual level. Estimation of causal effects in observational data requires adjustment for observed covariates typically via regression modeling or matching. For data with group level treatments, analysts often use multilevel regression models, but currently there are few options for matching with group level treatments. We develop an optimal matching algorithm based on dynamic integer programming that simultaneously considers balance at both the subject and group level. That is, for each treated group the algorithm searches for matched controls at both the group and subject level. The matching algorithm includes the option to use optimal subset selection, covariate constraints, and fine balance among others. We also demonstrate how estimation and sensitivity analysis may be conducted via randomization inference given the matched, grouped structure of the data. We provide an example of this multilevel matching algorithm with education data to assess the effect of switching to a private school from a public school.