Only four of the 13 leading agent gateways govern large-language-model calls, Model Context Protocol tools and agent-to-agent traffic natively, according to a MarkTechPost ranking published this week.
The list evaluates platforms that sit between autonomous agents and the models, tools and peer agents they call. A plain LLM gateway only routes inference requests; an agent gateway adds MCP tool governance, agent identity and, increasingly, A2A traffic control. MarkTechPost weighed five criteria: traffic coverage, identity and authorization, deployment options, maturity and price transparency.
Kong AI Gateway, Bifrost from Maxim AI, Gravitee AI Gateway and Prisma AIRS AI Gateway are the only products that score full marks across LLM, MCP and A2A, the report said. Kong moved AI features into first-class entities in version 2.0, though the 2.x architecture requires a Konnect control plane and its MCP OAuth2 plugin remains in tech preview. Bifrost, an Apache 2.0 Go binary, supports all three traffic types in a single deployment and claims 11 microseconds of mean overhead at 5,000 requests per second, a figure an independent review notes excludes HTTP handling and JSON marshalling.
Gravitee runs LLM, MCP and A2A proxies on one runtime with shared authentication, policy and OpenTelemetry chains, but its A2A proxy and agent management sit behind an Enterprise edition listed at $2,500 per month. Prisma AIRS, which reached general availability in July after Palo Alto Networks completed its Portkey acquisition in May, runs in SaaS and hybrid models with a unified enforcement point, though pricing is not public.
Cloud-native gateways trade integration for lock-in. Amazon Bedrock AgentCore Gateway offers a managed endpoint with semantic tool search, Cedar policies and on-behalf-of token exchange at $0.005 per 1,000 invocations, but A2A is HTTP passthrough only and search calls cost five times the invoke rate.
Microsoft’s Azure API Management extends its gateway to model APIs and MCP servers with A2A preview limited to JSON-RPC on v2 tiers. Google Cloud Agent Gateway assigns SPIFFE IDs to agents and can apply Model Armor, yet cross-cloud support is listed as coming soon and model routing and fallback are undocumented.
Arcade.dev occupies a different lane. It is an actions runtime, not a model router, and checks the overlap of user and agent permissions on every action while keeping credentials out of the LLM. The company lists 8,000-plus agent-optimized tools and charges $25 per month plus $0.10 per auth event and $0.01 per tool call, but it must be paired with an LLM gateway for model routing.

Open-source options target self-hosted and air-gapped deployments. IBM ContextForge federates MCP, REST, gRPC and A2A via Helm, though it has no published commercial support and IBM recommends upgrading to v1.1.1 after security bulletins covering v1.0.0 to v1.1.0.
Agent Router, the rebranded Envoy AI Gateway that joined the Agentic AI Foundation in September, offers an OpenAI-compatible API and MCP gateway on Envoy with A2A in preview and no built-in guardrails. Obot bundles MCP and LLM gateways with device monitoring and per-user auth, but self-hosting alone does not settle downstream authorization.
Composio provides a SaaS integration layer with managed OAuth and a free tier of 100,000 tool calls per month, though Arcade notes sessions are discoverable by default unless allowlisted. TrueFoundry combines an LLM gateway, centralized MCP access and MLOps tooling at $25 per user per month with 20,000 requests and tool calls per user, but its client-credentials MCP flow uses a shared identity rather than the user’s.

How the 13 gateways compare on protocol support and deployment
| Gateway | LLM | MCP | A2A | Open source | Deployment | Entry price |
|---|---|---|---|---|---|---|
| Kong AI Gateway | Yes | Yes | Yes | Gateway core only | Hybrid (Konnect CP) | Usage-based |
| Bifrost | Yes | Yes | Yes | Yes (Apache 2.0) | Self-hosted, VPC, air-gapped | Free OSS |
| Gravitee AI Gateway | Yes | Yes | Yes (Enterprise) | Not disclosed | Self-hosted, hybrid, SaaS | $2,500/mo |
| AgentCore Gateway | Yes | Yes | Passthrough | No | AWS managed | $0.005 / 1K calls |
| Arcade.dev | No | Yes | Not documented | Framework only | Cloud, VPC, air-gapped | Free; $25/mo |
| Prisma AIRS AI Gateway | Yes | Yes | Yes | No | SaaS, hybrid | Not disclosed |
| Azure API Management | Yes | Yes | Preview | No | Azure, self-hosted gateway | Not disclosed |
| Google Agent Gateway | Not documented | Yes | Yes | No | Google Cloud | Agent Compute usage |
| IBM ContextForge | Yes | Yes | Via MCP / passthrough | Yes (Apache 2.0) | Self-hosted (Helm) | Free |
| Agent Router | Yes | Yes | Preview | Yes (Apache 2.0) | Self-hosted (Kubernetes) | Free |
| Composio | No | Yes | Not documented | Not disclosed | Cloud, enterprise self-host | Free; $29/mo |
| TrueFoundry | Yes | Yes | Not documented | No | Cloud, on-prem | Free; $25/user/mo |
| Obot | Yes | Yes | Not documented | Yes | Self-hosted, Obot Cloud | Free (Community) |
Identity and enterprise features sit behind paid tiers
Most identity features — single sign-on, SCIM provisioning and token exchange — are gated behind enterprise licenses, the MarkTechPost analysis found. Bifrost reserves SSO, SCIM and token exchange for its Enterprise tier. Gravitee’s A2A proxy requires Enterprise Edition.
Azure API Management limits A2A to v2 tiers. Prisma AIRS and Kong do not publish pricing for the tiers that unlock full identity and audit depth. Arcade advises verifying delegated auth and audit depth for agent actions before committing.
What drives the choice
If agents only call models, an LLM gateway is enough. If they call tools, MCP governance is required. If they delegate to other agents, check A2A support carefully — preview, passthrough or JSON-RPC only outside the top four. Then check identity: ask whether user and agent permissions are enforced separately at execution time. Finally, check where the gateway must run: single cloud, your VPC, or air-gapped.
Founded in Lille, France, in 2014 by four co-founders building API strategies for French retailers, Gravitee operated profitably and entirely bootstrapped for seven years before taking its first venture capital, an $11 million round in July 2021.
The platform began as an open-source API manager under an Apache 2.0 license and later added an identity and access management module, positioning itself as a combined API and IAM platform used by enterprises including Tide and Auchan.
“Only Kong, Bifrost, Gravitee and Prisma AIRS govern LLM, MCP and A2A natively,” the MarkTechPost ranking concluded.
Worth a look