Cloud/Connectors/Audit Logs
Audit Logs
The connector's own trail: every request and connection made to it, filterable by type, tool, token, client and date, with errors split by Agent, Upstream and Vinkius causes.
Audit Logs is the connector's personal record. The subtitle: "Review all requests and connections made to this connector." Where the AI Governance reports aggregate your whole fleet, this tab is the single-connector ledger, one row per event.
Audit Logs
Review all requests and connections made to this connector.
| Time | Type | Client | Detail | Code | Agent | Upstream | Vinkius |
|---|---|---|---|---|---|---|---|
| 09:42:18 | Request | Cursor | create_issue | 200 | 0 | 0 | 0 |
| 09:41:55 | Request | Claude Desktop | search_code | 500 | 0 | 1 | 0 |
| 09:40:02 | Connection | Cursor | Connected | — | 0 | 0 | 0 |
| 09:38:47 | Request | VS Code Copilot | list_issues | 400 | 1 | 0 | 0 |
| 09:36:10 | Request | Cursor | get_file_contents | 502 | 0 | 0 | 1 |
| 09:30:00 | Connection | Windsurf | Disconnected | — | 0 | 0 | 0 |
Agent: 4xx, the agent sent a bad request · Upstream: 5xx, the MCP provider failed · Vinkius: 5xx thrown by Vinkius, SLA-relevant
The mockup is the real table. Switch the filters between All types, Requests and Connections and watch the stream change.
Two kinds of events
The trail records both things that happen to a connector: Requests, an agent calling a tool with its tool name and status code, and Connections, clients connecting and disconnecting. Each row shows the timestamp, the type, the client, the detail, and the code.
Filters that answer questions
Above the table sit the filters: All types / Requests / Connections, All tools, a Date range (Today, Last 7 days, Last 30 days), the Token used, the Client and the plan. Together they answer the everyday questions: what did this client call today, which tool is throwing, did this token get used after someone left.
The error split you can act on
For requests, errors are not one number. The table splits them by cause, and the column tooltips teach the difference:
- Agent: 4xx, the agent sent a bad request;
- Upstream: 5xx, the MCP provider failed;
- Vinkius: 5xx thrown by Vinkius, SSRF block or internal throwable, the SLA-relevant column.
That split is the difference between fixing your prompt, phoning the API vendor, and calling support. The same attribution powers the fleet-wide Request Failures report; here it is per request, in order.
Retention and streaming
History depth follows your plan, and the tab says it honestly: "Upgrade to Pro or Business for full audit retention." When you need the raw events in your own systems, Log Streaming forwards them as they happen; the summary numbers here, requests, errors, avg latency, top tools, are the same ones the console surfaces for the whole fleet in Tool Reliability.
Why this tab exists
When an agent did something at 2 a.m., this is the page that says exactly what: which client, which tool, which arguments, which outcome, and whose fault the failure was. It is the connector's black box, written down before anyone needs it.