Author here. I built Maverik because I want to be able to 1) assess + quantify improvements in agents 2) predict what an agent that would be exposed in a business setting (ie. a single agent for a specific task in a UI) is estimated to cost. So this is JMeter for agents :) be able to define test suites, pass criteria, run them and compare cold hard numbers.

An interesting find (intuitive but still interesting in terms of numbers), for one of my test agents, adding harmless MCP Servers that have tools this agent does not need, make it 28% more costly in terms of input+output tokens.

Anyway, next up for me are small improvements and maybe a crack at integrating the 2026-07-28 MCP Spec (stateless core, tasks, ...).