Academy
6. Auditing with the Execution Log
Use the executions history to review every tool call — identifying which tool fired, when, from which origin, and whether it succeeded or failed.
Every tool call the AI makes through your MCP connections is recorded in the Execution Log. This dedicated view captures four critical data points per entry:
• Tool — which specific tool was invoked (e.g., "create_issue", "send_email"). • Time — the exact timestamp of the call. • Origin — where the call was triggered from (a ChatPro conversation, an OUPI Studio agent run, a workflow step, a RAG import, or Deep Research). • Outcome — whether the call succeeded or failed.
Because connections work platform-wide, the log aggregates calls from every surface into a single audit trail. This is essential for debugging, compliance, and understanding how your AI agents interact with external services.
The origin field deserves special attention. A single MCP connection can be consumed by ChatPro, Studio, the Assistant, workflow steps, RAG imports, and Deep Research — all simultaneously. When something goes wrong, the origin tells you which surface triggered the problematic call. This lets you isolate issues to a specific agent run or workflow step rather than guessing across the entire platform.
Combine origin with outcome filtering to answer questions like: "Did my workflow's Slack tool calls all succeed last night?" or "Which Studio agent run produced a failed API call?"
When the log shows a failed outcome, the most common root cause is an expired or revoked OAuth authorization on the external service's side. The connection appears intact in the MCP Hub, but the token it holds is no longer valid.
To resolve this:
- Open the failing connection in the Connections view.
- Re-authorize it — this refreshes the OAuth grant.
- If the service changed its required permissions or scopes, grant them again during re-authorization.
If re-authorization doesn't help, delete the connection and recreate it. Remember: deleting removes it from every surface platform-wide, so coordinate with any workflows or agents that depend on it.
Cross-reference the execution log with tool inspection. Open a connection to see the exact tools it exposes — the same names appear in the log. This helps you confirm whether the AI called the right tool or picked an unexpected one.
Tool calls don't have a separate cost line. Their credits are included in the AI operation that triggered them — the conversation, agent run, or workflow. So when auditing spend, look at the originating operation's credit usage, not the execution log itself.
Open the Execution Log in the MCP Hub. Pick a recent failed entry (or any entry if all succeeded). Note the tool name, timestamp, and origin. Then navigate to the corresponding connection and verify the tool exists in its tool list. If the entry failed, try re-authorizing the connection and re-triggering the action.
The Execution Log is your single audit trail for every MCP tool call across the platform. Each entry records the tool, timestamp, origin surface, and success/failure outcome. Use origin to isolate issues to a specific agent or workflow. Failed calls usually mean expired OAuth — re-authorize or recreate the connection. Cross-reference tool names with the connection's tool list for verification, and remember that tool-call costs roll into the triggering AI operation's credits.