Skip to content

Security policy

mcpwn is an offensive security testing tool. It is designed to intercept and tamper with Model Context Protocol traffic so operators can assess agents and MCP servers they own or are explicitly authorized to test. Using it against systems without authorization may be illegal.

If you find a security issue in mcpwn itself (e.g. it leaks captured data, egresses traffic, or can be made to crash a host destructively), please open a GitHub security advisory or a private report rather than a public issue.

Responsible disclosure of findings made with mcpwn

Section titled “Responsible disclosure of findings made with mcpwn”

If you use mcpwn to find a vulnerability in a third-party MCP server or agent:

  1. Report privately to the maintainer of the affected project first.
  2. Allow reasonable remediation time (commonly 90 days) before publishing.
  3. Do not exfiltrate or retain any real user data encountered during testing.
  4. When you publish, a link back to mcpwn is appreciated but not required.

mcpwn keeps all captured traffic local. It has no telemetry and makes no outbound network connections of its own. Session logs (--log) are written only to the path you specify and may contain secrets observed on the wire - treat them accordingly.