Mongo: "The Monitoring agent is fairly robust, and MMS automatically sends an Agent Down alert when a Monitoring agent becomes unavailable. For most applications, a single Monitoring agent is sufficient. However, you can choose to run a second agent to provide redundancy. Larger numbers of agents do not provide significant benefit. See: Monitoring Architecture for information."

So we had these guys running a single agent that went down during Black Friday and their system went sideways. We had no time series data view into why their system failed.
We can recommend agent, you might be able to do it, but we need to recommend two. Even if the agent is perfect, machines fail, and usually under heavy load when you can least afford to miss monitoring stats.