Agentgateway

Agentgateway is a proxy that sits between agents and the things they talk to: LLM providers, MCP tool servers, other agents over the A2A protocol, and ordinary HTTP, gRPC, and TCP services. It puts all of that traffic in one data plane, so routing, authentication, policy, rate limiting, and telemetry are applied in one place instead of across separate gateways.

For LLM traffic it exposes a unified OpenAI-compatible API in front of providers such as OpenAI, Anthropic, Gemini, Bedrock, and Vertex, with budget and spend controls, prompt enrichment, load balancing, and failover. For self-hosted inference it routes to model servers such as vLLM, TGI, and Triton using Kubernetes Inference Gateway extensions, deciding on GPU utilization, KV cache, LoRA adapters, and queue depth.

For MCP it federates tools from multiple servers over stdio, HTTP, SSE, and Streamable HTTP transports, turns OpenAPI services into tools, and adds OAuth authentication. For A2A it handles capability discovery, modality negotiation, and task collaboration between agents built on frameworks such as LangChain, CrewAI, and ADK.

It runs standalone from a flat YAML configuration or on Kubernetes with a built-in controller and Gateway API support, and includes an admin UI for exploring agent-to-agent and agent-to-tool connections and testing routes. Agentgateway is a Linux Foundation project and has joined the Agentic AI Foundation.