MCP Aegis

Open-source security layer for the Model Context Protocol ecosystem.

MCP Aegis aims to make third-party MCP servers safer to discover, evaluate, install, and run. The initial direction is:

security-aware MCP subregistry
install-time verification and risk scoring
policy-based approval and version pinning
sandboxed execution for local MCP servers
runtime audit trails for MCP tool usage

Why This Project

The MCP ecosystem is growing quickly, but trust and runtime isolation are still weak in most real-world setups. MCP Aegis focuses on the missing control plane:

which MCP servers should be trusted
which versions are allowed
what a server can access at runtime
how teams can audit what happened later

Current Status

This repository is in the architecture phase.

Available now:

initial system architecture

Planned next:

threat model
policy spec
MVP service skeleton

Architecture

See docs/architecture.md for the current design, including:

high-level architecture
trust boundaries
install and runtime flows
deployment topology
OSS-friendly repository structure

MVP Direction

The first open-source release should focus on:

MCP subregistry with security metadata
scan pipeline and explainable risk scoring
policy engine for allow, block, and pin decisions
Docker-based sandbox for local stdio MCP servers
CLI installer for approved MCP configurations

License

Apache-2.0

MCP Servers