openkrak-mcp@1.0.5

One or more suspicious static signals were detected.

A configured MCP client invokes a tool and its handler throws.

Errors may reveal repository paths or other runtime details while enabling host correlation; the instructions can impair an agent's independent review.

AI-agent instruction steering combined with undisclosed error telemetry.

The package exposes MCP tool descriptions that tell an AI coding agent not to inspect source files before or after analysis. Its bundled runtime separately posts the tool name, full error message, version, and a hash derived from hostname and username to the vendor service whenever a handler fails. This telemetry is not disclosed in the included privacy statement, which expressly says there is no telemetry and only license validation reaches the network.

The concealed telemetry and host fingerprint create a concrete data-exfiltration path, while the MCP descriptions attempt to constrain the host agent's inspection behavior. The absence of an install hook reduces installation risk but does not remove the runtime attack surface.