Activity and usage
A team's call records in Open MCP Portal (each gateway's Activity tab and call pages) and the Usage dashboard.
Open MCP Portal records every list and call through a gateway as metadata: who called what, when, through which gateway and server, how it went, how long it took and how large it was. Arguments and results are never in call records; they are kept only where a team turned on recording call contents. initialize and ping aren't recorded.
Activity
A gateway's Activity tab lists its calls, newest first, for the team's owners and admins (members are told to ask one of them). Filter by outcome, service account and date. A Recorded mark means the call's contents were recorded; open the call to view them.
| Outcome | Means |
|---|---|
| OK | Answered |
| Tool error | The tool ran and reported an error |
| Denied | Refused by policy (no access, a role too low, the capability unavailable, a key limited to other gateways). The client was told it wasn't found |
| Upstream error | The server was unreachable, refused or answered something malformed |
| Timed out | The server didn't answer in time |
| Too large | The request or response was over 1 MiB |
| Bad request | The client's request was invalid |
| Needed a connection | Not sent: the person had to connect or reconnect an account first, and their client got a link |
| Rate limited | Not sent: a rate limit was exceeded |
Each row names the caller: a person with an API key ("API key: Laptop"), a person through a connected app ("via Claude Code", or "via <name> (unverified)" for an app that registered itself), or a service account with the key used ("Service account · platform-tools · Key: GitHub Actions").
A call's page
Opening a call shows when it happened, who made it, the method, the exposed name called, the gateway, the server, the outcome with its error class, the account it needed (if any), its duration and sizes, the client's name when it was known, and its Request ID (also in the portal's logs). If the call's contents were recorded, View contents shows them; see Recording call contents.
Usage
Usage in the team's sidebar (owners, admins and editors) summarises calls through the team's gateways, to its MCP servers and by its service accounts:
- Range: last 24 hours, 7 days, 30 days, 90 days, or custom dates (UTC), at most 90 days.
- Group by: gateway, server, capability, team, kind of caller (people with API keys, people with connected apps, service accounts), day or hour (hour for at most 7 days). There is no grouping by person.
- Totals for the range: calls, error rate, p50 and p95 latency, bytes sent to and received from servers.
- A chart of calls per day (or hour) by outcome, a latency chart, and the grouped table, most calls first.
Usage comes from hourly totals the portal computes every 5 minutes from the call records, so the newest calls show up within a few minutes. Latency percentiles are accurate to a bucket (5 ms up to 120 s). Times are in UTC.
How long records are kept
Call records are kept for 90 days by default, usage totals for 400; a platform admin can set other periods, and your team may have its own. A call involving two teams is kept for the shorter of their periods. Your Team settings › Limits & retention shows what applies. See Retention.