There are a few such tools, yeah.  There are some tools that will automatically dock your compound to a wide variety of known targets, though sadly I'm not familiar with them.  One that I *have* used is chemspider, and more particularly LASSO as implemented within it.  This essentially compares your molecule of interest to known drugs, and uses that information to suggest proteins it might bind to.  If you see a protein with known cancer implications, that could help.

All that being said, the value of virtual screening tends to be for libraries or for compound optimization.  Predictions on a single molecule are likely to be sketchy at best, and as likely a waste of time as not, especially given that it only takes a week or so to actually screen your molecule against actual cancer cell cultures and get a direct answer.

I think these tools of target fishing and inverse docking would help specially about therapeutic cancer targets -i assume you may know- as all of them have been implemented in cancer drug design research even for known compounds.

http://www.cbligand.org/TargetHunter/

http://59.78.96.61/pharmmapper/

http://www.dddc.ac.cn/tarfisdock/

http://www.dddc.ac.cn/pdtd/

 - there is also another interesting scheme have been implemented on screening NCI-60 cell lines and was further developed in this tool

http://crdd.osdd.net/oscadd/cancerin/

- this one also i think it may be interesting to you as it compiles a very efficient algorithm to detect compounds of anticancer activity from their SMILEs

http://bsb.kiz.ac.cn/CDRUG/

I´m worried that an anticancer activity is a feature too general and not so straightforward to predict. Cancer is a class of highly complex diseases and I´m not sure there are anticancer agents useful for general purpose.

For structure-based approaches you should have at least an idea of some possible targets or mechanisms of actions (eg. methylating agent, kinase inhibitors, tubulin inhibitors etc). You can find some QSAR models available but they are local which means that they are able to provide reliable results only in specific cases (eg. anti-breast cancer agents or for some specific chemical classes).

As already said, computational approaches applied to a single compound are unlikely to provide you meaningful results...