The MCP Maturity Model: From Chaos to Governance
Shannon Williams, co-founder of Obot AI and Rancher Labs, and former CNCF governing board member, delivers the keynote at the Agentic AI Foundation Summit. He walks through a year of helping companies make sense of MCP at scale: the shadow MCP problem, the real breaches that surfaced it, and the four-stage maturity model that maps how enterprises are climbing out of chaos. If you are responsible for AI adoption, governance, or platform security in 2026, this is the field report. The Shadow MCP Problem: why almost every enterprise already has MCP servers running it does not know about The Healthcare Breach Story: how one company discovered its MCP footprint only after data started flowing somewhere it should not One Year of MCP Attacks: the timeline from the first malicious lookalike MCPs last fall through reference-server vulnerabilities and upgrade-channel attacks The Four-Stage MCP Maturity Model: chaos, standardization, gateway and access control, and AI-driven governance Stage 2 Reality: enterprise agreements with Anthropic, OpenAI, and Cursor are how most companies are climbing out of shadow IT Stage 3 Patterns: MCP gateways behind proxies, identity-provider integration with Okta and Entra, granular per-tool access, full transaction audit Stage 4 Future: applying AI to the audit logs themselves so policy enforcement scales Beyond the Gateway: why local clients like Cursor, Claude Code, and stdio MCPs are now the next security frontier The Skills, Plugins, and Agents Explosion: how the MCP picture got messier in six months and what to do about it A 3-Step Action Plan: audit your environment, map your exposure, then apply governance Built for platform leads, security teams, and engineering executives trying to enable agentic AI without absorbing the blast radius. Links and Resources: Agentic AI Foundation: https://agenticaifoundation.com Obot AI: https://obot.ai Obot on GitHub: https://github.com/obot-platform/obot Shannon Williams on LinkedIn: / shannonwilliams2 Model Context Protocol: https://modelcontextprotocol.io Timestamps (approximate, may need adjustment): 00:00 Intro: Bangalore, MCP Dev Summit, and the year that was 01:30 What companies actually ask Obot in customer meetings 03:00 The shadow MCP problem and the healthcare breach 04:30 The MCP attack timeline of the last year 06:30 The four-stage MCP maturity model 07:30 Stage 1, shadow MCP: no visibility, downloaded locally 08:30 Stage 2, standardization: enterprise tool agreements 09:30 Stage 3, gateways: access control, audit, identity 10:30 Stage 4, AI-driven governance 11:30 Why the local client is the next security frontier 12:00 Three steps to start: audit, exposure, governance 12:30 Close and applause #MCP #ShadowAI #ModelContextProtocol

Keynote: After the AI Hype – What’s Real, and What’s Next - Richard Campbell - 2026

How GPT, Claude, and Gemini are actually trained and served – Reiner Pope

Customer Conversation: Data and AI Trust CONVERGE for the Agentic Era

Zig 2026: No-AI Policy, $670K Foundation, Left GitHub & Why Zig Isn’t 1.0 - Andrew Kelley Explains

Mitchell Hashimoto’s new way of writing code

MCP Gateways: Why They're Critical to AI Deployments

Billionaire's WARNING: I'm SELLING. The Crash Is Already Here!

Deep Dive into LLMs like ChatGPT

Inside Anthropic, the $965 Billion AI Juggernaut | The Circuit

A leader’s guide to advanced team structures in an agentic world | AWS Events

Open AI Is In Deep Trouble

Creator of C++: Bell Labs, Negative Overhead Abstraction, Mistakes | Bjarne Stroustrup

What AI Agent Skills Are and How They Work

Attacking AI - Jason Haddix - NDC Security 2026

OWASP's Top 10 Ways to Attack LLMs: AI Vulnerabilities Exposed

Wrong Reasons to Build an MCP Server (and What to Do Instead)

Stop zero days without a patch: You need to learn eBPF

Complete Backend Course | Build and Deploy Your First Production-Ready API

Andrej Karpathy: From Vibe Coding to Agentic Engineering w/ Stephanie Zhan

