On-Demand: MCP Security in the Enterprise

The Model Context Protocol (MCP) has become a foundational building block for agentic applications, enabling agents to securely discover and invoke tools.

In this workshop, you'll learn how to deploy and secure MCP servers in enterprise environments. We'll cover:

• MCP fundamentals and how agents connect to MCP servers
• Why MCP services should run as standalone processes using Streamable HTTP rather than stdio
• Securing MCP servers with OAuth and the MCP specification
• Decoupling authentication and authorization from tool implementations using a proxy architecture
• How agentgateway secures and proxies MCP servers, agents, and LLM workloads
• Enterprise SSO integration using JWTs issued by corporate Identity Providers (IdPs)
• Why enterprise JWT-based authorization is often a better fit than the specification's OAuth/OIDC model for internal services
• Emerging identity challenges, including capturing both the end-user identity and the identity of the agent acting on their behalf
• An end-to-end "On Behalf Of" scenario using Solo Enterprise for agentgateway and its Secure Token Service (STS)

By the end of the session, you'll understand how to build MCP deployments that meet enterprise requirements for security, identity, and governance.

After the workshop

Put what you learned into practice: see how agentgateway secures MCP servers, agents, and LLM traffic, then try the hands-on Federate MCP Servers with agentgateway lab.

Frequently asked questions

What is MCP security?

MCP security is the set of controls that protect Model Context Protocol (MCP) servers and the AI agents that call them: authenticating clients, authorizing which tools each agent or user can invoke, protecting credentials, and auditing every tool call. Because MCP gives agents direct access to tools and data, it needs the same identity and policy rigor as any production API.

How does agentgateway help secure MCP?

agentgateway sits in front of MCP servers as a proxy, so authentication, authorization, and observability are enforced at the gateway instead of being re-implemented in every tool. It validates JWTs from your identity provider, applies per-tool access policies, and records tool calls, while handling Streamable HTTP MCP servers, agents, and LLM traffic in one place.

SSO vs API keys for MCP servers?

Static API keys are simple but hard to rotate, scope, or tie to a real user. Enterprise SSO issues short-lived JWTs from your corporate identity provider, so every MCP request carries a verifiable user identity (and, with On-Behalf-Of patterns, the agent identity too) that the gateway can authorize and audit. For internal enterprise MCP services, SSO-based JWT authorization is usually the better fit.