API Gateway

CVE-2026-109492 in Microsoft's open-source MCP Gateway let any authenticated user who could create adapters or tools take over another user's workload or the built-in tool gateway through a name collision. Here is how the bug class works, the control-plane and per-tool authorization checks an MCP gateway needs, and a checklist for evaluating any gateway.

MCP servers are now operational control surfaces for cloud and API platforms. This article explains the trust model, control stack, and audit architecture enterprises need before AI agents can safely execute infrastructure actions.

The MCP 2026-07-28 spec hardened authentication and routing, but OAuth alone still cannot decide whether a specific tool call should run. This guide explains the authn-vs-authz split and shows how to enforce runtime policy on tools/call.

The fast-mcp-telegram and LiteLLM CVE chains show that authentication failures rapidly become unauthorized tool execution. The fix is fail-closed, runtime tool-call authorization at the MCP boundary.

Agent traces are observability signals, not accountability evidence. This guide defines a causal, replayable authorization commit log for MCP tool calls so teams can prove who delegated access, why policy allowed an action, and what happened next.

OpenAPI-to-MCP gateways can instantly expose REST endpoints as agent tools, but endpoint filtering alone does not enforce least privilege. This guide shows how to classify generated tools by risk, enforce runtime policy per delegator and intent, keep credentials server-side, and log auditable MCP decisions.

The LiteLLM CVE-2026-42271 and Starlette BadHost CVE-2026-48710 chain turned authenticated command injection into unauthenticated RCE. The deeper lesson: AI gateways hold model credentials, route sensitive traffic, and expose MCP utility endpoints — and need action-time authorization, not flat API keys.

MCP auth is necessary, but it is not the same thing as agent authorization. If you want secure agent systems, you need identity, delegation, policy, and runtime enforcement beyond OAuth.

An MCP proxy forwards traffic between MCP clients and servers. An MCP gateway adds identity, consent, authorization, and audit to every agent tool call.

A practical blueprint for securing Model Context Protocol (MCP) agents across identity, consent, policy, and audit layers without rewrites.

Announcing Permit MCP Gateway, a new trust and enforcement layer for MCP that brings identity, consent, fine-grained authorization, auditability, and runtime control to AI agent actions.

Discover best practices for authorization in REST API. Learn about API authorization layers, actors, tools like Permit.io and OPAL.

Kong is a popular API gateway, but managing access to its APIs and services is hard - especially when required advanced permissions models like RBAC/ABAC/ReBAC