MCP gives agents a clean way to call tools. Production teams need one more thing: a durable record of what happened.
Run logs are the difference between a clever demo and an operational system.
What a run log should answer
- Which user and workspace initiated the action?
- Which tool was called?
- What parameters were passed?
- What policy checks ran?
- What changed, if anything?
Without that record, a team cannot review, reproduce, or trust agent work.
Reads and writes are different
Read tools can usually run with lightweight approval. Write tools need stronger boundaries: validation, preview, approval, and rollback context.
That distinction should be visible in the run log, not hidden in implementation details.
The operator experience
The operator should be able to open a run, scan each step, and understand why the agent did what it did. If the result looks wrong, the team should know whether the issue came from user intent, tool input, policy, or source data.