Cloud/AI Governance/Server Traffic
Server Traffic
The Connectors side of the wire: which MCP servers carry your traffic, when demand peaks, where the milliseconds go and which server is about to become a problem, ranked and health scored.
Agent Activity showed the actors; Server Traffic shows the stage: every Connector (MCP server) serving your fleet, ranked by traffic, with its latency split into what the upstream API takes and what Vinkius adds, and a health score on every row. The console's own promise: "Spot the server that is about to become a problem before your users do. Fleet health, latency trends, and traffic hotspots in one read."
Server Traffic
Spot the server that is about to become a problem before your users do. Fleet health, latency trends, and traffic hotspots in one read.
Traffic Distribution
Requests by day of week × hourRequest Volume by Server
Total calls · color = latency health| MCP Server | Total Calls | API | Health |
|---|---|---|---|
| GitHubHigh Volume | 4,187 | 61ms | Healthy |
| SlackHigh Volume | 2,934 | 78ms | Healthy |
| NotionHigh Volume | 2,178 | 92ms | Healthy |
| Jira | 1,842 | 171ms | Moderate |
| Stripe | 1,264 | 55ms | Healthy |
| Linear | 986 | 71ms | Healthy |
| HubSpot | 743 | 198ms | Moderate |
| Vercel | 512 | 48ms | Healthy |
| Supabase | 387 | 122ms | Healthy |
| Asana | 214 | 702ms | Slow |
The mockup runs on the console's Sample Data. On a paid plan the same screen shows your live Connectors, per organization, with your real traffic.
The four counters
- Active Servers. How many Connectors carried traffic in the period: the live size of your fleet, not the installed size.
- Total Requests. Everything that crossed the proxy, across all Connectors.
- Avg Latency with a split bar: the emerald share is the upstream API's own time and the amber share is Vinkius overhead, milliseconds by design. This is the receipt that the control layer costs almost nothing.
- Peak Hour. The window with the highest traffic, computed from the period you are viewing. Capacity planning starts here.
Two charts: when the fleet works, and who carries the load
Traffic Distribution is a punchcard of requests by day of week × hour. It shows when your agents actually work: the hotspots are your operating rhythm. A bright cell at an unexpected hour is a story worth opening, because someone or something ran when nobody expected.
Request Volume by Server ranks Connectors by total calls, and the bar color is latency health: green for fast servers, amber for moderate, red for slow. The busiest server and the slowest server are different conversations, and this chart keeps them apart.
The server table: health, not hope
Every row is one Connector, and the console grades it instead of guessing:
- MCP Server, with a High Volume badge once a server passes 2,000 calls in the period, so the heavy hitters identify themselves.
- Total Calls per server.
- The latency split in three columns: API (the upstream's own time), Overhead (what Vinkius adds) and Total, color coded with the same thresholds everywhere: green under 200ms, amber under 800ms, red after.
- Req/Day, the per day pace of that server.
- Health as an explicit verdict: Healthy under 200ms, Moderate under 1s, Slow after that, with a colored dot you can scan down the column in one second.
Clicking a row opens that Connector's detail. The empty state is the console's own: "No MCP servers recorded in this period."
The two sides of every request
Agent Activity and Server Traffic are the same traffic seen from both ends: the client that asked and the server that answered. When Mission Control flags a latency or failure pattern, one screen tells you who asked and this one tells you what answered. The natural next stops are Tool Reliability (tool by tool) and Request Failures (when answers go wrong).