MCP server gate tests: every launch, counted and tested

When a major MCP server launches, ScriptMasterLabs runs the same two measurements on it. First, a capability count taken from the vendor’s own launch materials — how many tools, how many of them write, delete, or move money, and which controls the vendor actually names. Second, a live gate test: the launch’s characteristic actions are scored by the SML decision harness (local-heuristic-v1, calibrated=false) against the standard decision-gated bands — ≥0.80 auto-act, 0.50–0.79 advisory hold, <0.50 escalate. Every test below is dated, and every score on its page was produced live on that date.

The pattern being tested is decision-gated machine payments: decide first, pay — or act — second. The product form of the same check is the Spend Gate. Honest limits, stated once for the whole series: the harness is a heuristic decider, not a calibrated model, and some launches expose so little write surface that the correct finding is “nothing to gate” — those pages say so.

All 34 tests, newest first

New launches are added as they ship and pass the same two measurements. If a server’s vendor corrects a capability count, the page is corrected and the correction is noted on the page.

SCRIPTMASTERLABS · READ · EXPLORE · VERIFY · EARN

Books, free tools, sources & affiliate opportunities

Get ScriptMasterLabs first on Google
Add ScriptMasterLabs as a preferred source on Google