This very high end GPU cracker performs 200GH/s:
That's 200 billion hashes per second (2e11 hashes/s). With a search space of 1e19 possible numbers, that means it will take ~500,000 seconds (about 1.5 years) to exhaustively search parameter space. Of course on average you'll find a match in half of that time.
In short, MD5 is a fast algorithm, which normally means brute force is easy. However in this case the large size of your search space negates a lot of that.
Depending on what happens for a successful user though, MD5 still might not be the best choice. A top end hashing rig is too slow by a factor of ~200,000. That seems like a lot, but you're basically asking "How long until my hardware might be 200,000x faster?". While the growth pace of computer hardware has slowed, it's not unrealistic to expect that gap to shrink substantially in just a few years.
Finally, you also have to be concerned about weaknesses in pre-image resistance. I haven't "checked in" on MD5's status in a while, but last I heard MD5 was still pre-image resistant. Other aspects of MD5's cryptographic security have been broken (aka it is no longer collision resistant, although that is not relevant here). Therefore, while MD5 is still pre-image resistant, that could change any day. Without pre-image resistance an attacker might come up with a technique to find a payload that gives the same hash as your number, without having to actually brute-force the number.
In short, you are safe at the moment. The safety margin though is perhaps not that high. Whether or not you need to take action depends on what you are securing on the other side of this.