Runtime enforcement where your agents already pass

Keep your AI gateway.
Add the decision it can't make alone.

Your agents already reach their tools through a gateway. MeetLoyd plugs into it as its authorizer. Every call is checked against the agent's identity, its stop switch, the tools it was granted and its budget, and the decision is signed into the audit trail. You keep the gateway you run today.

12 gateways
Cloud, open source and edge
One decision
The same rules, whichever gateway asks
Fails closed
No answer means no access

Works with

Each gateway asks in its own way. MeetLoyd answers all of them with the same decision. Where the gateway passes the request body on, the tool named in the MCP request itself is what gets checked, not only the headers.

Cloud

Google Agent Gateway

Authorization extension, request and content profiles.

Checks the MCP body
Cloud

AWS AgentCore Gateway

Request interceptor, ready to deploy.

Checks the MCP body
Cloud

Amazon API Gateway

Lambda authorizer, ready to deploy.

Headers only
Cloud

Azure API Management

An inbound policy you paste in.

Checks the MCP body
Edge

Cloudflare Workers

A Worker in front of your MCP server.

Checks the MCP body
Open source

Envoy & Envoy Gateway

Native external authorization.

Checks the MCP body
Open source

Istio

Custom authorization provider; the mesh identity of the calling agent is recognised.

Checks the MCP body
Open source

agentgateway

External authorization, gRPC or HTTP.

Checks the MCP body
Open source

kgateway

External authorization through its gateway extension.

Checks the MCP body
API management

Kong Gateway

Through its policy-agent plugin.

Checks the MCP body
Proxy

Traefik

Forward authentication middleware.

Checks the MCP body
Proxy

NGINX

Subrequest authorization.

Headers only

Using a gateway that isn't listed? If it can call an external authorizer, it can call MeetLoyd. Tell us which one.

What is decided on every call

Identity

Which agent is calling

Its own key, a federated token, or the identity your gateway or mesh already established. Agents nobody registered don't get through.

Stop

Whether it is still allowed to run

Stop an agent, a team or a whole organisation, and every gateway refuses it within about a second. No redeploy.

Grants

Whether it may use this tool

Tools are denied unless granted, and a grant can be limited by time, network or declared purpose. Delegated calls carry who delegated them.

Budget

Whether it can still spend

A budget that is exhausted blocks the next call, at the gateway, before the cost is incurred.

Every decision, allowed or refused, lands in a signed audit trail you can export to your SIEM.

Each gateway proves who it is

A decision is only as good as the gateway that asked for it. Each gateway is registered, bound to one organisation and revocable on its own: with a certificate we issue for it, or with a token when the gateway can't present one. A gateway that can't prove itself is not listened to, and it can't vouch for an agent.

Runs near your agents

The decision is made in your region, from a continuously updated copy of your policy, so the gateway isn't waiting on a round trip across the world.

Fails closed

If the decision can't be made in time, or the policy copy falls behind, the call is refused rather than waved through.

Bring the gateway you run

Tell us which gateway sits in front of your agents. We'll connect it in a working session in your environment and show you the first refused call in the audit trail.

Book a call →

Product names are trademarks of their respective owners. Listing a product means MeetLoyd works with it; it does not imply endorsement or partnership.