Agent Trust Lab

Independent evidence for how reliably AI agents select MCP tools, with frozen benchmark results, preserved semantic misses, multi-batch reliability evidence, and independent MCP evaluation.

Documentation

Trust profiles forAgents & MCP Servers.

Agent Trust Lab turns public artifacts, reproducible benchmark results and verified operational evidence into version-bound Trust Profiles. Selection evidence is published as a separate badge class; it never silently changes transport evidence, ownership or static profile grade.

10 Trust Profiles

7 MCP Servers

1 Agent demo

37.8% Evidence coverage · 34/90 fields

Evidence coverage is not a quality score. Missing evidence is shown as NOT_PROVIDED, never converted to zero. Selection badges use a separate S0–S4 evidence policy and do not improve static E0–E4 evidence grade or ranking.

Machine-readable by default

Open the catalog API →

What a Trust Profile means

Read the methodology →