I’m an independent researcher (and professionally, the Global Director of Research at Reink Media) looking for an endorsement for the cs.AI (Computer Science/Artificial Intelligence) category on arXiv.

The Context I didn't start by writing a paper; I started by building a system. Over the last year, I developed a production-grade Model Context Protocol (MCP) server for the forex market. It’s currently live with 45 distinct tools and renders 28 different dynamic widgets across Claude, ChatGPT, and a custom web app.

The Paper The paper is titled Protocol-Constrained Agentic Systems: A Neuro-Symbolic Architecture for Hallucination-Resistant Financial Execution.

My core argument is that in high-stakes domains like finance, we cannot rely on LLMs to be "smart enough" to avoid critical errors. Instead, I propose an architecture that uses MCP as a "hallucination firewall." This strictly decouples the probabilistic layer (the LLM parsing intent) from the deterministic layer (the tool executing the trade). It effectively treats the protocol schema as a type system for agent actions, guaranteeing by construction that invalid tool calls cannot reach the execution layer.

You can read the full paper and see the architecture in the PDF here: https://www.stevenhatzakis.com/research/protocol-constrained-agentic-systems

The Request If you are a registered endorser for cs.AI and find this work relevant, I would appreciate your support so I can submit this paper on arXiv.org.

Endorsement Code: LZRTFH Link: https://arxiv.org/auth/endorse?x=LZRTFH

Thanks for your time.

Steven Hatzakis