13 Agent gateways to watch in 2026 for MCP, A2A, and LLM traffic : Kong, Bifrost, AgentCore, Arcade

13 Agent gateways to watch in 2026 for MCP, A2A, and LLM traffic : Kong, Bifrost, AgentCore, Arcade
2 min read

AI agents no longer just call a model: they use tools via the MCP protocol, delegate tasks to other agents via the A2A protocol, and act within corporate systems. Agent gateways sit on this path and decide who can do what. A ranking published by MarkTechPost compares 13 based on traffic coverage, identity, deployment, and price.

An agent gateway is a control point that authenticates agents, filters the tools they see, and logs every call. A classic LLM gateway is limited to routing requests to language models. The agent gateway adds governance for MCP tools, agent identity, and increasingly, control of A2A traffic. The ranking is based on five criteria: traffic coverage, identity and authorization, deployment options, maturity, and price transparency.

In first place is Kong AI Gateway, which governs LLM, MCP, and A2A traffic from a single control plane. Its version 2.0 placed AI functions as dedicated entities: providers, models, MCP servers, and agents. The 2.x architecture requires a Konnect control plane, and Arcade reports that the OAuth2 plugin for MCP remains in Tech Preview.

Bifrost, developed by Maxim AI, is a Go-based gateway distributed under the Apache 2.0 license. According to Maxim, it covers over 10,000 models from more than 25 providers. Its benchmark claims an average overhead of 11 µs at 5,000 RPS, but an independent review notes this figure excludes HTTP handling and JSON serialization. SSO, SCIM, and token exchange are reserved for the Enterprise edition.

Major cloud offerings

Amazon Bedrock AgentCore Gateway offers a managed access point for tools, models, and other agents, with semantic tool search and Cedar policies. Its pricing starts at $0.005 per 1,000 invocations. It is exclusive to AWS, and the A2A protocol is handled as simple HTTP forwarding. Search calls cost $0.025 per 1,000, five times the invocation rate, according to Maxim.

At Microsoft, the Azure API Management AI Gateway extends the service to model APIs, MCP servers, and A2A agent APIs. A2A support is in preview, limited to JSON-RPC and v2 offerings. Google Cloud Agent Gateway, for its part, assigns a SPIFFE ID to each agent and can apply Model Armor to requests and responses. Maxim indicates that multi-cloud compatibility is ‘coming soon.’

Prisma AIRS AI Gateway, from Portkey, also illustrates the sector’s consolidation. Palo Alto Networks finalized the acquisition of Portkey in May 2026, and the gateway has been generally available since July 2026. Its pricing is not public.

Choosing a gateway

The ranking recommends starting with traffic: if agents only call models, an LLM gateway is sufficient; if they use tools, MCP governance becomes necessary; if they delegate to other agents, A2A support must be carefully checked. Next, verify if user permissions and agent permissions are applied separately at runtime. The deployment location comes last: a single cloud, a private VPC, or an isolated environment.

The ranking highlights that most identity functions, such as SSO or token exchange, are reserved for enterprise licenses.

S’abonner
Notification pour
guest

0 Commentaires
Le plus ancien
Le plus récent Le plus populaire
0
Nous aimerions avoir votre avis, veuillez laisser un commentaire.x