aegisproxy
v0.1.4
Published
Autonomous security layer for AI agents — Claude Desktop integration
Downloads
20
Readme
aegisproxy
Security proxy for AI agents — Claude Desktop integration.
Wraps Claude Desktop's MCP servers with AegisProxy's threat detection, audit logging, and policy enforcement.
Requirements
- Node.js 18 or higher
- Claude Desktop installed
- An AegisProxy account (aegisproxy.com)
Setup
npx aegisproxy init
Follow the prompts to authenticate and configure. Claude Desktop will be updated automatically.
Commands
aegisproxy init — First-time setup wizard aegisproxy status — Check current config and connection aegisproxy --serve — Start the MCP proxy server (Claude Desktop calls this automatically — you don't need to run it manually)
How it works
aegisproxy sits between Claude Desktop and its MCP tool servers. Every tool call Claude makes passes through a two-tier inspection engine before being forwarded:
- Tier 1 (local, sub-millisecond): URL allowlist, path traversal, known-bad patterns — no network call
- Tier 2 (cloud, ~100ms): prompt injection, slopsquatting, semantic analysis — only for ambiguous cases
All decisions are logged to your AegisProxy dashboard in real time.
Security
Your API key is stored in ~/.aegisproxy/config.json with 600 permissions (readable only by you). It is never passed to Claude Desktop or stored in claude_desktop_config.json.
