Agent Identity for MCP: Prefactor’s Approach to Secure, Auditable AI Agents (Demo + Deep Dive)

Boards want “an agent strategy”—security teams want control. In this talk and live demo, Matt Doughty (CEO, Prefactor) and co-founder Simon unpack why traditional OAuth isn’t enough for the Model Context Protocol (MCP) and how agent identity changes the game for enterprise security, governance, and observability.

𝗪𝗵𝗮𝘁 𝘆𝗼𝘂’𝗹𝗹 𝗹𝗲𝗮𝗿𝗻:

Why MCP introduces a new attack surface—and where guardrails are missing today

The difference between human identity vs. agent identity (and why it matters)

How Prefactor authenticates, secures, and audits agents with signed payloads and immutable logs

Internal vs. external MCP use cases (workforce agents vs. customer-facing servers)

Embedding security into chat UIs and plans to integrate with LibreChat and OpenWebUI

SaaS today, with on-prem/private cloud on the roadmap

Enterprise patterns: workload identity, task-based access control, and removing end-user consent friction

𝗪𝗵𝗼 𝘁𝗵𝗶𝘀 𝗶𝘀 𝗳𝗼𝗿:

Security leads, platform teams, and MCP builders who need centralized identity, authorization, and audit for agents—without slowing down delivery.

𝗞𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀

  • Treat agents as first-class identities separate from humans and clients
  • Push authorization decisions to MCP servers with verifiable context
  • Gain real-time visibility via session management + audit trails
  • Prepare for enterprise MCP with provisioned, policy-driven access—not ad-hoc API keys

If you’re exploring MCP for regulated industries, or scaling servers internally, this session will help you design a secure, observable, enterprise-grade foundation for agents.