Overview
The Analytics dashboard gives you aggregate insights about your MCP server usage. Unlike Traffic Logs (which show individual requests), Analytics shows trends and summaries over time.
Access account-level analytics from Analytics in the dashboard sidebar. Per-server analytics are available on the Analytics tab of any server detail page.
Time range presets
All charts and KPI metrics update to reflect the selected time range:
| Preset | Data shown |
|---|
| 24H | Hourly breakdowns for the past 24 hours |
| 7D (default) | Daily breakdowns for the past 7 days |
| 30D | Daily breakdowns for the past 30 days |
| 90D | Daily breakdowns for the past 90 days |
Account-level analytics
The account-level view aggregates across all your servers.
KPI cards
Four headline metrics shown at the top:
| Metric | What it measures |
|---|
| Total Requests | Cumulative tool calls in the selected period, with % change vs prior period |
| Error Rate | Percentage of calls that resulted in an error — green below threshold, red above |
| Avg Latency | Mean execution time across all tools and servers, with % change vs prior period |
| Active Servers | Count of servers that received at least one request in the period |
Daily (or Hourly) request trend
An area chart showing total requests and errors over time.
- 24H preset: hourly data points
- All other presets: daily data points
Hovering over the chart shows the exact values for that day or hour. Use this to spot sudden spikes, drops, or error surges.
Latency trend
A multi-line chart showing:
- Avg latency (blue) — mean execution time
- P95 latency (yellow dashed) — 95th percentile; the worst case for 95% of requests
- P99 latency (red dashed) — 99th percentile; the worst case for 99% of requests
The Y-axis scales automatically — values under 1000ms are shown in ms; values over 1000ms are shown in seconds.
Watch the P95 line, not just the average. A rising P95 with a flat average means a subset of requests are getting slow — often a database query missing an index or a rate-limited upstream API.
Top servers
A horizontal bar chart showing your top 8 most-called servers by request count.
Geographic distribution
A horizontal bar chart showing the top 10 countries by request origin (ISO 2-letter country codes). Useful for understanding where your users are and for diagnosing region-specific latency issues.
Server-level analytics
Click any server from the Top servers chart, or open a server’s Analytics tab, to see analytics scoped to that server.
Server KPI cards
| Metric | What it measures |
|---|
| Total Requests | Calls received by this server in the period |
| Error Rate | Percentage of calls that failed |
| Avg Latency | Mean execution time for this server’s tools |
| Max Concurrency | Peak number of simultaneous active sessions |
Request trend
Same dual-line area chart (requests + errors) as account level, scoped to this server.
A horizontal bar chart showing the distribution of calls across this server’s tools — up to the top 10.
Use this to:
- Find tools with zero calls (bad description? unused feature?)
- Understand the AI’s behaviour: a tool called unexpectedly often may have a too-broad description
Session trend
An area chart of unique client sessions over time. A session is a single AI client connection.
- Rising sessions = more clients connecting → growing adoption
- Flat sessions + rising requests = each session is more active → more intensive use per user
Traffic by hour
A bar chart showing average request count by hour of day (0–23, in your local timezone). Use this to understand peak usage hours and plan maintenance windows.
Using analytics effectively
Identify optimisation targets:
Look at P95 latency. If it’s over 3–5 seconds for any time period, your tools are too slow for a good user experience. Check the tool breakdown to find the culprit, then investigate query indexes, pagination, or caching.
Find unused tools:
In the tool breakdown, zero-call tools over 30 days are candidates for cleanup or improved descriptions. The AI may not be calling them because the description doesn’t match what users ask for.
Debug sudden error spikes:
Find the time window when errors spiked in the trend chart, then open Traffic Logs filtered to that window to see what went wrong.
