Task completion
Did the intended application state actually change?
WebMCPify Bench · Coming beta
Compare agent tasks with evidence.
A planned benchmark for measuring whether agents can discover, invoke and verify real website capabilities—across repeatable scenarios and explicit safety boundaries.
NOT AVAILABLE YET Scenarios and scoring are in development. No benchmark scores are published or implied.
What Bench will measure
Success is more than a green tool response.
Bench is designed to compare end-to-end task outcomes with the evidence needed to reproduce and challenge them.
Discovery quality
Did the agent find the relevant capability and understand its schema?
Safety behavior
Were authentication, approval and risk boundaries respected?
Verification strength
Was success checked independently of the agent’s own report?
Reliability
Can the result be repeated across runs and environments?
Efficiency
How much time and agent work did a verified outcome require?