---
title: "What is an MCP gateway? How it differs from an AI gateway"
description: "An MCP gateway is a proxy between AI agents and many MCP servers that centralizes authentication, tool access, and audit logs. How it works and vs AI gateways."
type: "glossary-term"
canonical_url: "https://theaiengineering.com/glossary/mcp-gateway"
markdown_url: "https://theaiengineering.com/glossary/mcp-gateway.md"
topic: "Model Context Protocol"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-11-11"
keywords: ["MCP gateway","MCP gateways"]
citation: "The AI Engineering Editors. \"What is an MCP gateway? How it differs from an AI gateway.\" The AI Engineering, updated Nov 11, 2026. https://theaiengineering.com/glossary/mcp-gateway"
---

# What is an MCP gateway? How it differs from an AI gateway

Also called MCP gateways.

> **Definition of MCP gateway:** An MCP gateway is a proxy that sits between MCP hosts and many MCP servers, exposing them through one endpoint and centralizing authentication, tool access policy, and audit logging.

An MCP gateway is a proxy that sits between MCP hosts, such as coding agents and chat apps, and an organization's [MCP servers](https://theaiengineering.com/glossary/mcp-server), and exposes those servers through one managed endpoint. The [Model Context Protocol](https://theaiengineering.com/glossary/model-context-protocol) specification does not define gateways; they are an operational pattern. To a host, a gateway looks like one server, but it implements few or no tools of its own. It authenticates callers, decides which tools each user or agent may see, forwards each call to the server that owns the tool, and logs every invocation.

## How an MCP gateway works

An MCP gateway works as a policy point on the tool path. The host connects once, the gateway collects the tool lists of its registered servers, filters them for the caller, and routes each `tools/call` to the right backend with that server's credential attached. Every call passes through one place, so audit logs and rate limits apply to all servers.

**Figure 1: Hosts see one endpoint; the gateway applies policy and routes each call.**

Two MCP hosts connect to one MCP gateway, which applies authentication, tool filtering, and audit logging before routing calls to three backend MCP servers.

Components:

- **Coding agent**: MCP host
- **Chat app**: MCP host
- **MCP gateway**: Auth · tool allowlist · audit log
- **Code host server**
- **Database server**
- **Ticketing server**

Groups:

- Backend MCP servers: Code host server, Database server, Ticketing server

Connections:

- Coding agent → MCP gateway: one endpoint
- Chat app → MCP gateway
- MCP gateway → Code host server
- MCP gateway → Database server
- MCP gateway → Ticketing server

The specification anticipates such intermediaries. Its [tools page](https://modelcontextprotocol.io/specification/2026-07-28/server/tools) says proxies that aggregate several servers should resolve tool-name collisions, for example by prefixing names with a server identifier, and allows `tools/list` to vary with the authorization on the request, which is how per-user filtering works. The [Streamable HTTP transport](https://theaiengineering.com/glossary/mcp-transport) mirrors the method and tool name into `Mcp-Method` and `Mcp-Name` headers so a gateway can route without parsing JSON. List results carry a `cacheScope`: a shared gateway may cache and reuse a `"public"` response for any user, but must not share a `"private"` one across authorization contexts.

Some AI gateways double as MCP gateways. [Bifrost](https://docs.getbifrost.ai/mcp/overview), an open source AI gateway, exposes the servers it connects to through one `/mcp` endpoint that hosts such as Claude Desktop and Cursor can use, filters the tool list per virtual key, and does not execute tool calls on its own unless an operator turns on its agent mode. Model calls and tool calls then share one set of keys, policies, and logs.

Other open source examples include [Docker MCP Gateway](https://docs.docker.com/ai/mcp-catalog-and-toolkit/mcp-gateway/), which runs each server in an isolated container and injects its credentials; IBM's [ContextForge](https://ibm.github.io/mcp-context-forge/latest/), which federates MCP servers, A2A agents, and REST or gRPC APIs behind one endpoint; Microsoft MCP Gateway, a reverse proxy and management layer for MCP servers on Kubernetes; and agentgateway, a Linux Foundation project that proxies LLM, MCP, and A2A traffic. The [MCP platforms directory](https://theaiengineering.com/platforms/mcp-platforms) lists more.

## MCP gateway vs AI gateway

An [AI gateway](https://theaiengineering.com/glossary/ai-gateway) governs traffic from applications to models; an MCP gateway governs traffic from agents to tools. Both centralize authentication, policy, and logs. An AI gateway meters tokens, routes between model providers, and caches completions, while an MCP gateway controls which tools each caller can discover and call and holds the credentials for each backend server.

| | AI gateway | MCP gateway |
|---|---|---|
| Sits between | Applications and model providers | MCP hosts and MCP servers |
| Request format | Usually an OpenAI-compatible model API | MCP JSON-RPC messages |
| Main policy | Model access, token budgets, rate limits | Tool allowlists and per-user tool filtering |
| Credentials held | Provider API keys | One credential per backend server |
| Typical log entry | Model, tokens, latency, cost | Tool, caller, arguments, result status |

The two categories overlap. Some products proxy both kinds of traffic, and several gateways in the [AI gateways directory](https://theaiengineering.com/platforms/ai-gateways) include MCP features. A gateway is also a natural place to pin approved tool definitions and block a tool whose description changes, a defense against [tool poisoning](https://theaiengineering.com/glossary/tool-poisoning). Like any shared proxy, it concentrates credentials and becomes a dependency for every agent, so it needs replicas and its own secret management.

[What is Model Context Protocol](https://theaiengineering.com/articles/what-is-model-context-protocol) explains where gateways fit in the protocol, and [what is an AI gateway](https://theaiengineering.com/articles/what-is-an-ai-gateway) covers the model-side pattern.

## Related terms

- [MCP server](https://theaiengineering.com/glossary/mcp-server): An MCP server is a program that exposes tools, resources, and prompts to AI applications through the Model Context Protocol, running locally over stdio or remotely over Streamable HTTP.
- [AI gateway](https://theaiengineering.com/glossary/ai-gateway): An AI gateway is a proxy layer between applications and model providers that exposes one API for many models and centralizes routing, failover, access control, cost tracking, caching, and logging.
- [Model Context Protocol (MCP)](https://theaiengineering.com/glossary/model-context-protocol): Model Context Protocol (MCP) is an open protocol, introduced by Anthropic in November 2024, that standardizes how AI applications connect to external tools, data sources, and prompts through MCP servers.
- [MCP transport](https://theaiengineering.com/glossary/mcp-transport): An MCP transport is the binding that carries Model Context Protocol messages between a client and a server; the specification defines stdio for local subprocesses and Streamable HTTP for remote servers.
- [Tool poisoning](https://theaiengineering.com/glossary/tool-poisoning): Tool poisoning is an attack in which a malicious or compromised MCP server hides instructions in a tool's description or schema, which the model reads as trusted context and may follow.
- [Guardrails](https://theaiengineering.com/glossary/guardrails): Guardrails are programmatic checks placed around a large language model that validate, filter, or block inputs, outputs, and tool calls so an AI application stays within its safety, security, and policy limits.

## Guides that cover mcp gateway

- [What is the Model Context Protocol (MCP)? How it works](https://theaiengineering.com/articles/what-is-model-context-protocol): The Model Context Protocol (MCP) is an open standard for connecting AI apps to tools and data. Learn its architecture, transports, tool calls, and security.
- [How to build an MCP server: a Python and TypeScript tutorial](https://theaiengineering.com/articles/how-to-build-an-mcp-server): Learn how to build an MCP server with the official Python and TypeScript SDKs: tools, resources, transports, testing, client config, and production security.
- [What is an AI gateway? How it works and when you need one](https://theaiengineering.com/articles/what-is-an-ai-gateway): An AI gateway is a proxy between your apps and model providers. Learn how it works, its core features, AI gateway vs API gateway, and open source options.
- [Bifrost vs LiteLLM: architecture, features, and performance compared](https://theaiengineering.com/articles/bifrost-vs-litellm): Bifrost vs LiteLLM compared: a Go single binary against a Python proxy, with deployment, routing, virtual keys, caching, MCP, benchmarks, and migration.

Part of the [Model Context Protocol](https://theaiengineering.com/topics/mcp) topic hub.

## Cite this definition

The AI Engineering Editors. "What is an MCP gateway? How it differs from an AI gateway." The AI Engineering, updated Nov 11, 2026. https://theaiengineering.com/glossary/mcp-gateway

- Title: What is an MCP gateway? How it differs from an AI gateway
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-11-11
- URL: https://theaiengineering.com/glossary/mcp-gateway
- Markdown: https://theaiengineering.com/glossary/mcp-gateway.md
