Errors & Latency

Error rate chart

A time-series line chart showing the percentage of tool calls that returned an error, plotted over time.

A tool call is counted as an error when:

• The status field is "error" in the tool call record
• The proxy received a non-2xx response from the upstream server
• The proxy hit a timeout or connection error

What to watch for:

• Sudden spikes → server deployment or config change broke something
• Steady baseline above 0% → persistent issue worth investigating
• Error rate correlating with high latency → the server is overloaded

Latency histogram

A histogram of tool call durations (ms) over the selected time window. Shows the distribution of response times, with p50, p95, and p99 percentile markers.

Interpreting the histogram:

• Bimodal distribution → two classes of tool calls (fast vs slow)
• Long tail to the right → occasional very slow calls; may indicate timeouts or large payloads
• p95 creeping up over time → gradual performance regression

Per-server breakdown

Both charts support filtering by server using the server selector. Use this to:

• Isolate whether an error spike is coming from one server or all servers
• Compare latency profiles across server versions or environments