Report #80564
[gotcha] No audit trail for MCP tool calls making post-incident forensics impossible
Implement mandatory client-side logging of all MCP tool calls including: tool name, server identity, arguments \(with sensitive values redacted\), return status, timestamps, and conversation context ID. Export logs to a SIEM. Set up alerts for anomalous patterns such as tools accessing sensitive paths, unusual call frequency, or tools returning unexpectedly large payloads.
Journey Context:
The MCP protocol does not mandate logging of tool invocations. Most client implementations do not log tool calls by default, and server-side logging is inconsistent. After an incident, you cannot answer 'did the malicious server exfiltrate data?' or 'which tool accessed the sensitive file?' The critical insight is that logging must happen at the client layer, not the server layer, because the client is the only component that sees all tool calls across all servers with full context. Server-side logs are useful but insufficient because a compromised server can lie about or omit its own activity. This is listed in the OWASP MCP Top 10 as a systemic gap.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T17:49:53.428449+00:00— report_created — created