Kong vs Solo.io agentgateway: platform feature or standalone proxy
Updated 19 November 2026 · Editorial assessment
At a glance
| Attribute | Kong | Solo.io agentgateway |
|---|---|---|
| What it is | Kong's Agent Gateway is built directly into Kong AI Gateway and requires no separate infrastructure. | agentgateway is an open source HTTP and gRPC gateway that handles ordinary application traffic and AI-native protocols in one data plane. |
| LLM, MCP and A2A | Kong AI Gateway lets you define a single endpoint for LLM, MCP or A2A traffic. | agentgateway offers native protocol support for MCP and agent-to-agent traffic. It can route to OpenAI, Claude, Gemini or any self-hosted model, with failover. |
| A2A policy and audit | Kong applies centralized authentication, authorization, and rate limiting to A2A traffic. It keeps tamper-evident audit trails of every A2A RPC call, including caller identity and capabilities invoked. | agentgateway routes agent-to-agent invocations between LangChain, CrewAI, ADK and your own runtime, with identity and tracing on every hop. Its policy layer comes with tamper-evident audit logs. |
| MCP security | Kong's open source repository lists MCP traffic governance, MCP security and MCP observability, plus MCP autogeneration from any RESTful API. | agentgateway can scope which tools each identity may invoke after authenticating callers into MCP servers. |
| Deployment | Kong's Agent Gateway runs wherever Kong AI Gateway runs, which Kong lists as on-prem, any cloud or hybrid configurations. | agentgateway lists bare metal, VMs, containers, and Kubernetes as deployment targets, with xDS-based updates without service interruption. |
| Managed and commercial options | Kong offers Dedicated Cloud Gateways, where Kong hosts the control plane and data planes for you, including scaling and upgrades. | agentgateway has an enterprise build, Solo Enterprise for agentgateway, which adds a hardened distribution, full-lifecycle MCP security and governance, and 24x7 support. |
| License | Kong Gateway's open source repository is licensed under the Apache License, Version 2.0. Kong Inc. also offers commercial subscriptions that add gateway functionality, commercial support and the managed Konnect control plane. | agentgateway operates under Apache 2.0 licensing. |
| Kubernetes fit | Kong documents a Helm install of Kong Gateway that pairs a Konnect control plane with a data plane deployed in your cluster. | agentgateway implements the Kubernetes Gateway API with support for HTTPRoute, GRPCRoute, TCPRoute and TLSRoute. |
How does each one coordinate agents?
A gateway applies policy to the calls that pass through it. Deciding which agent takes a task belongs to the runtime or collaboration layer above it, so read this section as how each product handles agent calls on the wire.
In Kong, A2A calls run through the same AI gateway that governs LLM and MCP traffic, and the gateway can capture detailed telemetry on every A2A call. In AI Gateway 2.x you create AI Model Provider and AI Model entities instead of configuring the AI Proxy plugin directly, then attach AI Policies for guardrails, transformations and rate limiting.
agentgateway was designed as a unified gateway control plane and proxy data plane for HTTP, gRPC, MCP, A2A and LLM traffic. For A2A it supports capability discovery, modality negotiation, and task collaboration. Policy can take the form of allow/deny rules over method, path, headers and JWT claims evaluated per hop, or the decision can be delegated to your own service over gRPC or HTTP. On Kubernetes, the control plane translates Gateway API and custom resources such as AgentgatewayPolicy into proxy configuration.
When should you choose Kong?
- Your platform team already runs Kong AI Gateway, and Kong says you can drop Agent Gateway into your existing deployment to govern A2A calls.
- Data sovereignty requirements apply, and Kong pitches its hybrid model, which lets you keep data plane traffic on-prem while using Konnect for control plane management, at organizations balancing those requirements with operational simplicity.
- If finance needs chargeback, Kong says its usage tracking supports showback and chargeback across LLM, agent, and MCP usage. You can also set user, model, and time-bound quotas for LLM consumption and token spend that the gateway enforces.
- When governance has to cover the data agents read, Kong says its platform extends governance to how agents consume context from APIs and event data through Kong Context Mesh.
When should you choose agentgateway?
- A foundation-hosted project fits your sourcing policy: agentgateway joined the Agentic AI Foundation as a hosted initiative under the Linux Foundation.
- If your clusters standardize on Gateway API, agentgateway's docs describe deploying and dynamically configuring it on Kubernetes with a built-in control plane and Gateway API.
- Small pilots can run agentgateway either as a standalone binary or in Docker, on a local machine or a server.
Can you use them together?
You can split one estate between them. Kong would sit in front of the APIs, LLM calls and A2A traffic it already governs, and agentgateway in front of the MCP servers and agent routes you point it at, each with its own policy and audit trail. Plan how identity claims and traces line up across both before you write routing rules.
Both sit beside a runtime or collaboration layer that assigns work to agents. See agent mesh vs orchestration for where that split falls, and the protocols page for MCP and A2A background.
What should you watch out for?
- The Kong Operator route for installing AI Gateway 2.x on-prem on Kubernetes is labelled Tech Preview. Check that status against your production requirements.
- Konnect-managed AI Gateway runs data plane nodes in your environment that connect to Konnect for configuration and observability. Confirm whether that dependency suits your design or whether you need the self-hosted stack.
- agentgateway's README says the project is currently in active development and asks users to open GitHub issues for missing features.
- The standalone mode is described as perfect for development, testing, and simple deployments. Check whether that scope matches your production plans.
What do enterprise buyers ask?
Is agentgateway open source?
Yes. Its README calls it an open source proxy built on AI-native protocols, and the project operates under Apache 2.0.
Does Kong handle A2A traffic?
Yes. Kong says its AI Gateway governs agent-to-agent traffic with centralized authentication and authorization, observability and auditability.
Can either gateway expose REST APIs as MCP tools?
Both can. Kong AI Gateway can automatically generate MCP tools and servers on top of Kong-managed APIs, and agentgateway lets you drop in any OpenAPI spec and expose it as MCP tools.
Where do I start with identity requirements?
Write them per hop before you pick a gateway. The agent identity and access guide covers the questions to ask, and the RFP checklist turns them into requirements.
Which sources does this page cite?
- Agent Gateway: Agent-to-Agent Governance & Security (konghq.com)
- agentgateway (agentgateway.dev)
- Kong AI Gateway (docs.konghq.com)
- Kong/kong: README.md (GitHub)
- agentgateway Joins AAIF as an Open Gateway for Agentic AI Infrastructure (agentgateway.dev)
- agentgateway: AI Agent Gateway for LLM & MCP (solo.io)
- How to: Install Kong Gateway in Konnect with Helm (docs.konghq.com)
- Announcing A2A, MCP, and Kubernetes Gateway API support (agentgateway.dev)
- Designing agentgateway: A Unified High-Performance Gateway for AI and API Traffi (agentgateway.dev)
- agentgateway/agentgateway: README.md (GitHub)
- Overview (agentgateway.dev)
- Secure, Scalable AI Gateway for AI Connectivity (konghq.com)
- Documentation (agentgateway.dev)
Reviewed Nov 2026