1.9 KiB
Security Policy
Supported code
Security fixes target the current main release line. Historical design-pack, work-packet and experiment branches are retained for traceability and are not separate supported release channels.
Reporting vulnerabilities
Report suspected vulnerabilities privately to the repository owner. Do not publish access tokens, authorization headers, private infrastructure details, production configuration, audit data containing identifiers, signing material or exploit-sensitive evidence in a public issue.
A useful report includes the affected commit/component, minimal reproduction conditions using synthetic data where possible, expected and observed behaviour, impact, and whether the issue affects authentication, authorization, project/environment isolation, connector boundaries, secret redaction, Docker broker isolation, MCP framing/session handling or deployment controls.
Control-plane boundary
The canonical MCP Hub is a privileged operational control plane. Preserve least-privilege connectors, explicit read/write separation, repository and workspace allowlists, optimistic locking, validation gates, attributable changes and fail-closed behaviour. Public examples or SDK material must never include operational credentials, private topology or production connector configuration.
Security invariants
The normative security requirements in docs/02-security/SECURITY_INVARIANTS.md are release requirements. Contributions must preserve deny-by-default authorization, bounded target-specific connectors, secret non-disclosure, auditability, and the prohibition on unrestricted shell/filesystem/SQL/Docker capabilities through MCP.
Never commit live .env files, tokens, private keys, production secrets or unredacted sensitive runtime evidence.
Disclosure
Coordinate remediation and disclosure with the repository owner before publishing technical details that would materially increase exploitation risk.