Building the governance layer for production AI agents.
I'm Marco Arras. I ship the security, observability, and policy controls that move AI agents from chat windows into real infrastructure. Solutions Architect by day; independent founder building Cordon and Agent Toolbelt.
What I'm shipping
Cordon
Policy enforcement layer between AI agents and tool servers. Human-in-the-loop approvals, SQL-aware policies, closed-world tool catalogs, and full audit logging. Works with Claude Desktop, Cursor, and Windsurf.
getcordon.com →Agent Toolbelt
Modular MCP tools for AI agent workflows. Per-call billing, hosted on Railway. The package suite I build the rest of my infrastructure on top of.
agenttoolbelt.live →Build & Ship MCP Tools
Technical course. 7 modules, 33 lessons. Module 0 free.
ET Acquisition Agent
Autonomous STR deal analysis. 4 MCP servers + Cordon.
Water-Lemon
Hyper-local DFW real estate market data micro-SaaS.
ShowKit
No-backend binder generator for real estate listing agents.
About
The biggest blocker to enterprise AI adoption isn't LLM intelligence. It's safety and orchestration. That's the gap I'm building into.
I'm a Solutions Architect at Clearview Energy in Dallas, with ten years of full-stack TypeScript and Node delivery behind me. Outside my day job I ship independent products: Cordon, Agent Toolbelt, a technical course, and a handful of smaller bets in real estate and personal tooling.
Self-taught engineer with a humanities background (BA in Spanish Literature, Mizzou). I think about agents as systems that operate inside human policy rather than around it, which is part of why I ended up building a security gateway instead of yet another agent framework.
Writing & signals
Let's talk.
Moving agents from a chat window into production infrastructure? That's the conversation I want to have.
arras.marco@gmail.com →