When Agent A reports Server X responded in 120ms, that helps Agent B decide whether to use Server X. Thousands of agents reporting millions of interactions create trust scores no individual agent could produce alone.
Static scores don't have this property. GitHub stars don't improve because more agents use a server.
What We Built
The Dominion Observatory indexes 22,561 MCP servers across 22 categories. Only 117 (0.5%) have any behavioral record, and just 18 (0.08%) are independently verifiable at runtime. That gap is the whole point: a static catalogue says a server exists, it cannot say the server still works when your agent calls it. It's free, open, and available as both MCP server and REST API:
-
MCP endpoint:
https://dominionobservatory.com/mcp -
Trust check:
GET /api/trust?url=[server-url] -
Report interaction:
POST /api/report -
Leaderboard:
GET /api/leaderboard -
Compliance export:
GET /api/compliance
Check a server's trust score before calling it. Report outcomes after. The more agents participate, the more reliable scores become — for everyone.
The MCP ecosystem needs runtime trust. Static scores got us this far. Behavioral data takes us the rest of the way.
Dinesh Kumar builds infrastructure for the AI agent economy from Singapore. GitHub: vdineshk
Top comments (0)