Runlayer

AARM Extended

Enterprise control plane for MCP servers, skills, and agents

Overview

Runlayer is the enterprise control plane for MCP servers, skills, and agents. It gives organizations a single place to host, govern, and secure the AI tools their employees rely on — across clients like Cursor, Claude Code, ChatGPT, and VS Code — without forcing users to change their workflows. The platform combines a curated catalog of vetted MCP servers with real-time threat detection, identity-aware access control, and full observability across every agent action.

Classification

Coverage surface: MCPEndpointSaaSCloudData/DBAPI
Stage: Launched
Type: Commercial
Target audience: Enterprise
Deployment: SaaSSelf-hostedHybrid

Technical profile

Spec-grounded axes, verified by the TWG.

Interception architecture (R1): Protocol GatewaySDK Instrumentation
Policy model (R3): Hybrid
Authorization decisions (R4): ALLOWDENYMODIFYSTEP_UPDEFER

Conformance review

`R1`	Pre-execution interception	Pass
`R2`	Context accumulation	Pass
`R3`	Policy evaluation with intent alignment	Pass
`R4`	Five authorization decisions	Pass
`R5`	Tamper-evident receipts	Pass
`R6`	Identity binding	Pass
`R7`	Semantic distance tracking	Pass
`R8`	Telemetry export	Pass
`R9`	Least privilege enforcement	Pass

Platform capabilities

MCP gateway that proxies every tool call with real-time policy enforcement
Shadow detection and enforcement at the endpoint level via MDM and EDR
Purpose-built ML threat detection models trained on MCP-specific attacks
Deep identity integration with Okta, Entra, SSO, and SCIM
Tamper-evident audit logs of every agent action and tool call
Slack-based human-in-the-loop approval workflows for sensitive actions
Semantic distance tracking across long agent task horizons (R7)
Telemetry export to Splunk, Datadog, Honeycomb, and S3 (R8)
Least-privilege credential and tool scoping at execution time (R9)

Architecture

Runlayer enforces AARM requirements through two complementary interception patterns. The primary gateway pattern sits between AI clients and MCP servers, proxying every tool call through a control plane where policies and ML security models evaluate the request before it reaches downstream systems. A secondary shadow pattern extends visibility and enforcement to endpoints, so activity originating on local developer machines is brought under the same policy surface. Context accumulation is handled through full session reconstruction across clients and tools, giving the platform end-to-end visibility into agent intent and actions across the full lifecycle. Authorization decisions, identity bindings, and policy evaluations are written to an append-only audit layer.