perimeter security proxy for MCP-based AI agents. Intercept, audit, and block dangerous tool calls.
462

MASTYF.AI — PERIMETER SECURITY FOR AI AGENTS
Open-source (MIT) runtime security proxy for MCP. Sits between your AI agent and every MCP server, intercepting all tool calls and blocking malicious actions before execution.
WHAT IT BLOCKS (18 attack classes):
304-entry adversarial corpus validates every rule.
ARCHITECTURE: 6-Phase Defense Fabric
Lifecycle → Pre-Guard → Hooks → Policy Engine → Semantic → Spend
QUICK START
Open mode (no auth — local dev): docker run -p 4000:4000 -e DASHBOARD_AUTH_DISABLED=true rudraneel93/mastyfai:latest
Dashboard at http://localhost:4000 — no login required.
Auth-enabled mode (production):
docker run -p 4000:4000
-e DASHBOARD_AUTH_DISABLED=false
-e DASHBOARD_JWT_SECRET="$(openssl rand -hex 32)"
rudraneel93/mastyfai:latest
First visit creates admin account.
FEATURES
SDKs & PACKAGES
npm: npm install @mastyf_ai/langchain-middleware npm: npm install @mastyf_ai/openai-agents-middleware Python: pip install mastyf-ai Docker: docker pull rudraneel93/mastyfai:latest
ENVIRONMENT VARIABLES
DASHBOARD_AUTH_DISABLED true Set false to enable auth DASHBOARD_JWT_SECRET — Required when auth is enabled DASHBOARD_PORT 4000 Dashboard HTTP port MASTYF_AI_DB_PATH ~/.mastyf-ai/history.db MASTYF_AI_BLOCKING_MODE block block or flag (audit-only)
LINKS
GitHub: github.com/mastyf-ai/mastyf.ai Cloud: mastyf-live.vercel.app npm: npmjs.com/org/mastyf_ai PyPI: pypi.org/project/mastyf-ai ghcr.io: ghcr.io/mastyf-ai/mastyf.ai:latest
TAGS: latest, main LICENSE: MIT
Content type
Image
Digest
sha256:dd187b08b…
Size
341.2 MB
Last updated
about 2 months ago
docker pull rudraneel93/mastyfai