MCP Security: Keep Your AI Agents from Spilling the Tea
summary
Demo runs on Azure using KAgent/KMCP to self‑host models (Ollama, OpenAI via Azure) and expose controlled MCP tools; shows practical RBAC patterns, an RFC 8693 token exchange flow and an elicitation fallback that prompts for scoped tokens rather than handing over universal keys. Kubernetes pods isolate agents, Kyverno policies enforce tool and network limits, OIDC gates the UI, and an AI firewall blocks inverted prompt injection. Live demo proves the proxying model works and reduces accidental privilege escalation.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:08What is MCP security
- 4:59MCP security layers
- 8:55Token exchange vs elicitation
- 12:46MCP vs REST access
- 14:34MCP server pitfalls
- 20:36Live demo: K agent
- 31:04Security features demo
- 36:45Summary
- 37:09Q&A
related talks
29mMCP Security: Keep Your AI Agents from Spilling the Tea
50mMCP at Scale: Governing AI Agent APIs When Autonomous Systems Are Your Consume…
32mAI Agents with Keycloak in MCP
15mWhy Security Matters: The Risks of Agentic AI and How to Mitigate Them
27mKeeping Your Agents in Check: Layered Security for Agentic Platforms in Production
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
25mWho Let the Agents Out? Securing AI Workflows the Right Way
18mThe "Agentic Organisation": Governing Multi-Agent Developer Teams with Nutanix Agent Gateway and MCP