What is an MCP Gateway?
An MCP gateway is a service that sits between MCP clients and multiple upstream MCP servers, providing a single point for authentication, policy enforcement, audit logging and tool filtering across all of them. Unlike a proxy in front of one server, a gateway manages an organisation's whole MCP estate through one ingress.
WHY IT MATTERS
Each MCP server an organisation adopts brings its own credentials, its own tool surface and its own risk profile. Connecting AI clients directly to each server means N separate trust decisions, N credential stores and no shared record of what agents actually did. A gateway collapses that into one controlled path: clients connect to the gateway, and the gateway holds the upstream connections.
Centralising the connection point makes several controls practical that are otherwise scattered or absent:
- Authentication — one identity layer for people and agents, rather than per-server API keys pasted into client configs.
- Policy — every tool call can be evaluated by a policy engine before it reaches the upstream server.
- Audit — a complete, uniform record of requests and decisions across all servers.
- Tool filtering — exposing only an approved subset of upstream tools to each client or person.
The distinction from an MCP proxy is scope: a proxy fronts a single connection or server, while a gateway is fleet-wide infrastructure — the MCP analogue of an API gateway. Gateways often also act as aggregators, multiplexing many upstreams into one endpoint.
HOW POLICYLAYER USES THIS
PolicyLayer is a hosted MCP gateway and control plane. Teams register their upstream MCP servers, define deterministic policies, issue per-person scoped tokens, and point clients such as Claude Code, Cursor and Codex at the PolicyLayer gateway. Every tools/call is evaluated against policy — allow, deny, or log — before it executes, and every decision lands in the audit trail.