Next Generation Agentic Proxy for AI Agents

Agentgateway is an open source proxy designed for AI-native protocols, specifically MCP and A2A. It provides essential features such as security, observability, and governance for various forms of agent communication, including agent-to-LLM, agent-to-tool, and agent-to-agent interactions across diverse frameworks and environments.

Key features include:

• LLM Gateway - Routes traffic to major LLM providers like OpenAI and Anthropic through a unified OpenAI-compatible API, offering budget controls, prompt enrichment, load balancing, and failover capabilities.
• MCP Gateway - Connects LLMs to tools and external data sources via MCP, supporting tool federation and various transport methods including stdio, HTTP, and OAuth authentication.
• A2A Gateway - Facilitates secure agent-to-agent communication with features like capability discovery and task collaboration.
• Inference Routing - Intelligent routing to self-hosted models using Kubernetes, optimizing based on GPU utilization and queue depth.
• Guardrails - Implements multi-layered content filtering and moderation through various tools and custom webhooks.
• Security & Observability - Offers robust security measures including JWT, API keys, and fine-grained RBAC, along with OpenTelemetry for metrics and logs.

Agentgateway is suitable for developers and organizations looking to enhance their AI agent communication capabilities while ensuring security and observability.