Report #13980
[gotcha] MCP has no mandatory logging of tool invocations—malicious or misused tools operate invisibly
Implement client-side telemetry that logs every tool call with server identity, tool name, full arguments, and return value; feed logs into a SIEM or audit system; set alerts for suspicious patterns \(e.g., file reads of credential paths, outbound HTTP calls with large payloads, unexpected tool call sequences\); never rely on MCP servers to self-report their activity.
Journey Context:
The MCP specification defines how tools are called and how results are returned, but it does not mandate any logging, auditing, or telemetry. If a compromised tool reads sensitive files and exfiltrates data, there is no built-in record that it happened. Developers assume their application logging covers this, but MCP tool calls happen through a separate protocol layer that standard APM tools do not instrument. By the time you discover a breach, you have no forensic trail of which tools were called, with what arguments, or what they returned. This is listed explicitly in the OWASP MCP Top 10 because it is a systemic gap, not an implementation oversight.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T20:19:16.810388+00:00— report_created — created