Agent Gateway

Download the latest standalone binary for your platform from the GitHub releases page. The binary includes both the control plane and the proxy data plane.

# macOS (arm64)
curl -LO https://github.com/agentgateway/agentgateway/releases/latest/download/agentgateway-darwin-arm64
chmod +x agentgateway-darwin-arm64
mv agentgateway-darwin-arm64 /usr/local/bin/agentgateway

Create a YAML config file that defines a listener, an LLM backend, and a route. This example routes traffic to OpenAI's API through the gateway.

# agentgateway.yaml
listeners:
  - name: default
    protocol: HTTP
    port: 8080
backends:
  - name: openai
    type: llm
    provider: openai
    auth:
      apiKey:
        value: "${OPENAI_API_KEY}"
routes:
  - name: default
    backends:
      - openai

Run agentgateway with your configuration file. The gateway starts listening on port 8080 and proxies LLM and MCP traffic according to your routes.

OPENAI_API_KEY=sk-your-key agentgateway --config agentgateway.yaml

Add an MCP backend to your config to route tool calls through the gateway. The gateway supports stdio, HTTP, SSE, and Streamable HTTP transports for MCP servers.

# Add to agentgateway.yaml
mcpBackends:
  - name: my-mcp-server
    transport: http
    address: "http://localhost:9000"

Configure your AI client or MCP host to connect to agentgateway's listener instead of directly to the LLM or MCP server. The gateway transparently proxies and enforces all policies.

{
  "mcpServers": {
    "agentgateway": {
      "url": "http://localhost:8080"
    }
  }
}

For production Kubernetes deployments, use the official Helm chart. The chart installs the gateway controller and configures it to use Kubernetes Gateway API resources.

helm repo add agentgateway https://agentgateway.github.io/agentgateway
helm install agentgateway agentgateway/agentgateway \
  --namespace agentgateway --create-namespace