MCP Server Trust Scores
Real-time reliability data for Model Context Protocol servers, based on production monitoring across thousands of AI agents.
Monitor your own MCP serversAll tracked servers
Search by name or click column headers to sort.
No MCP server data yet
Trust scores appear once enough production traffic has been collected. Scores require at least 10 calls per server.
Methodology
How Trust Score is calculated
Each server starts at 100. Points are deducted for these signals:
Error rate
up to −40
A server with a 100% error rate loses 40 points. 10% errors → −4 points.
High latency
up to −20
P95 over 2s loses 20 points. P95 over 500ms loses 10 points. Under 500ms is free.
Retry loops
up to −30
Traces stuck calling the same tool over and over. A reliability red flag — 100% loops loses 30 points.
Scores are calculated from anonymized production data across all TraceHawk users — no organization, trace, or prompt content is exposed. Servers with fewer than 10 calls in the last 30 days are excluded to reduce noise. Updated every hour.
Want detailed MCP monitoring for your agents?
TraceHawk gives you per-server health dashboards, tool-level latency breakdown, and retry loop detection — for every MCP server your agents touch.