Audit tool invocations without storing unnecessary personal data

Audit evidence should answer who invoked which version, for what tenant, with what outcome and receipt. Copying full prompts and records into every log increases privacy and security risk.

Inspect a minimal event

The MCP contract lab creates an authored audit record.

from mcp_cases import audit_case

result = audit_case()
assert result["required_present"] is True
assert result["unnecessary_personal_fields"] == []
assert result["raw_arguments_stored"] is False
print(result["event"])

It includes time, tool/version, user pseudonymous ID, tenant, argument digest, outcome, latency and request ID. No raw rows, email, token or prompt are stored.

Match evidence to purpose

Define incident, access review, billing and debugging needs, then keep only required fields. Separate operational logs from controlled evidence stores when raw content is occasionally necessary. Apply access, encryption, retention and deletion.

Use stable reason codes and opaque receipts. Hashes can still be sensitive if inputs are guessable; add keyed designs or protected mappings where appropriate. Never log credentials.

Test observability privacy

Inject personal data and secrets into nested arguments/errors and scan exported traces. Verify tenant boundaries, reviewer access and retention deletion. Confirm enough metadata remains to investigate duplicate writes.

The Generative & Agentic AI course balances auditability with data minimization in MCP and LLMOps.

Exercise

Write an audit schema for five incident questions. Remove every field without a named purpose, then test investigation of a duplicate write using receipts and digests only.

Continue learning

This article is part of the MCP and integration contracts sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Design an approval-required write tool.
• Continue with MCP versus a direct API integration: evaluate the trade-off.

Reference: NIST Privacy Framework.