AgentGateway
AgentGateway is an open-source, AI-native proxy and gateway for routing, observing, and governing traffic to and from AI agents, LLM providers, and MCP servers. Built on the A2A and MCP protocols, it provides a unified gateway for LLM consumption, MCP tool federation, agent-to-agent communication, security, and observability. AgentGateway supports multi-provider LLM routing across OpenAI, Anthropic, Google Gemini, AWS Bedrock, and Azure OpenAI with built-in RBAC, JWT authentication, rate limiting, and OpenTelemetry integration.
APIs
AgentGateway
AgentGateway provides AI-native gateway capabilities for routing LLM traffic, federating MCP tools, enabling agent-to-agent communication, and applying security and observabilit...
Features
Routes traffic to OpenAI, Anthropic, Google Gemini, AWS Bedrock, and Azure OpenAI through a unified API with model aliasing, failover, and load balancing.
Connects LLMs to tools via Model Context Protocol with static and dynamic routing, tool federation, and stateful MCP sessions.
Enables secure, governed communication between AI agents using the A2A protocol for multi-agent orchestration.
Intelligently routes requests to self-hosted models based on GPU utilization and request priority.
Provides JWT, OAuth2, API key management, CORS, CSRF protection, MCP authentication, and external authorization support.
Supports request routing and matching, header manipulation, rate limiting, retries, gRPC routing, traffic splitting, and direct responses.
Integrates with OpenTelemetry for metrics, traces, and access logging with a built-in Admin UI and debugging tools.
Applies prompt guards, content filtering, regex filters, moderation policies, and custom webhooks for AI safety.
Tracks budget and spend limits per user, team, or application with RBAC-based controls on LLM consumption.
Supports prompt templates and enrichment for standardizing and augmenting requests before routing to LLM providers.
Use Cases
Route requests across multiple LLM providers with a single API, enabling failover, load balancing, and cost optimization without changing client code.
Aggregate tools from multiple MCP servers behind a single gateway endpoint, enabling agents to discover and invoke tools from any connected MCP server.
Apply organization-wide security policies, rate limits, budget controls, and content filters to all AI agent traffic through a centralized gateway.
Convert existing REST APIs into MCP-native tool endpoints that AI agents can discover and invoke through the Model Context Protocol.
Enable secure agent-to-agent communication using the A2A protocol, allowing specialized agents to delegate tasks to each other through the gateway.
Collect unified telemetry across all AI agent and LLM interactions to monitor cost, latency, and behavior at scale.
Integrations
Route to OpenAI GPT models through the AgentGateway LLM backend with model aliasing and budget controls.
Connect to Anthropic Claude models via the unified LLM gateway with failover and load balancing.
Route traffic to Google Gemini models through the AgentGateway multi-provider backend.
Integrate with AWS Bedrock for managed LLM access via the AgentGateway routing layer.
Route requests to Azure-hosted OpenAI models through the unified gateway API.
Connect to locally hosted Ollama models for self-hosted inference routing.
Route to vLLM inference servers with GPU utilization-aware routing for optimal performance.
Export metrics, traces, and logs to any OpenTelemetry-compatible observability backend.
Deploy and configure AgentGateway on Kubernetes using the standard Gateway API for dynamic configuration.