Skip to main content

​
What is Agentgateway?

Agentgateway is an open source data plane optimized for agentic AI connectivity. It provides drop-in security, observability, and governance for agent-to-agent (A2A) and agent-to-tool (MCP) communication — working within or across any agent framework or environment. Written in Rust for maximum performance, Agentgateway supports leading interoperable protocols including Model Context Protocol (MCP) and Agent2Agent (A2A).

Quickstart

Get Agentgateway running in minutes with a basic MCP proxy setup

Installation

Install from source, pre-built binaries, or Docker

MCP Proxy Guide

Learn how to proxy MCP servers with security and observability

A2A Guide

Set up agent-to-agent communication with the A2A protocol

​
Key features

MCP & A2A Protocol Support

First-class support for Model Context Protocol (MCP) and Agent2Agent (A2A) — the leading standards for agentic AI connectivity.

Security First

Built-in RBAC, JWT authentication, mTLS, and AI prompt guard to secure every agent interaction.

High Performance

Written in Rust from the ground up — designed to handle any scale with minimal latency overhead.

Dynamic Configuration

Zero-downtime configuration updates via xDS. Local file-watch reloads also supported.

Kubernetes Native

Built-in Kubernetes controller with Gateway API support for cloud-native deployments.

OpenAPI to MCP

Transform legacy REST APIs into MCP resources automatically using OpenAPI specifications.

Full Observability

OpenTelemetry tracing, Prometheus metrics, and structured logging out of the box.

Multi-Tenant

Support multiple tenants each with their own resources, routes, and policies.

​
How it works

Agentgateway sits between AI agents and their tools or other agents, acting as a transparent proxy that adds security, observability, and governance to every interaction. 
AI Agent → Agentgateway → MCP Server (tools, resources, prompts)
                       → A2A Agent (agent-to-agent tasks)
                       → HTTP/REST API (via OpenAPI transformation)
The gateway is configured through a declarative YAML/JSON configuration that defines:
  • Binds — ports the gateway listens on
  • Listeners — virtual hosts and protocols per port
  • Routes — traffic routing rules and matching criteria
  • Backends — upstream MCP servers, A2A agents, or HTTP services
  • Policies — authentication, authorization, rate limiting, and traffic shaping

​
Protocol support

ProtocolDescriptionTransport
MCPModel Context Protocol for AI toolsSSE (/sse), Streamable HTTP (/mcp), Stdio
A2AAgent2Agent for agent interoperabilityHTTP
HTTPGeneric HTTP/HTTPS trafficHTTP/1.1, HTTP/2

​
Community

Agentgateway is a Linux Foundation project.

Build docs developers (and LLMs) love

Get started for freeTalk to us