Please report security issues privately via Telegram @dracoeques rather than opening a public issue. You will get an acknowledgement within a few days and a fix or mitigation plan.

mcpkit runs tools on behalf of a language model, so the threat model centres on untrusted tool inputs. The network, filesystem and shell tools are disabled by default and each has documented guardrails — see docs/security.md.

The main branch receives security fixes. Tagged releases are patched on a best-effort basis.