Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Do you need an AI Gateway?

Do you need an AI Gateway?

LLM applications increasingly rely on MCP servers for tools, data, and automation, but teams often hit real-world snags: trusted server discovery, inconsistent authentication, lack of built-in security features, unreliability of tool calls, tool pollution, lack of proper observability, etc. All these issues have a significant impact on the developer experience and make it hard to properly operationalize MCP for real-world production environments.

In this talk, we’ll frame these pain points from a developer and platform persona angles, and discuss how an MCP gateway can help address them and provide a solid foundation to leverage MCP at scale in enterprise environments.

Avatar for Ignasi Barrera

Ignasi Barrera

October 02, 2025
Tweet

More Decks by Ignasi Barrera

Other Decks in Technology

Transcript

  1. Do you need an AI Gateway? 2 October | London

    IGNASI BARRERA Founding Engineer @ Tetrate
  2. ➜ Founding Engineer @ Tetrate ➜ ASF proud Member ➜

    Envoy AI Gateway contributor MCP DEV SUMMIT | EUROPE
  3. MCP DEV SUMMIT | EUROPE Some problems are well-known •

    Trusted server discovery • Authentication / Authorization • Server resiliency: high availability, load balancing, failover • Observability Some are new • Context bloating • Tool confusion • Tool poisoning, security guardrails • Decision fatigue
  4. MCP DEV SUMMIT | EUROPE Some problems are well-known •

    Trusted server discovery => MCP Registry (https://github.com/modelcontextprotocol/registry) • AuthN/ AuthZ => SEP-1036 (https://github.com/modelcontextprotocol/modelcontextprotocol/issues/1036) • Server resiliency: high availability, load balancing, failover • Observability => OpenInference (https://arize-ai.github.io/openinference/spec/) Some are new => Require deep understanding of the protocol • Context bloating • Tool confusion • Tool poisoning, security guardrails • Decision fatigue
  5. MCP DEV SUMMIT | EUROPE Envoy AI Gateway • Reliable

    Foundation with Envoy Proxy Built upon Envoy's battle-tested proxy technology to provide a stable, performant, and production-ready foundation that enterprises can rely on • Resilient Connectivity Across LLM Providers and Self-Hosted Models • Comprehensive Observability for Performance and Cost Management • Enterprise-Grade Security Features • Extensible Architecture