Operant AI

Overview

Operant AI provides runtime protection for AI agents, MCP servers, and agentic applications. The platform centers on two enforcement components: the Operant Endpoint Protector, which intercepts MCP tool calls, prompts, and shell executions before execution, and the Operant Agent Protector, which extends real-time enforcement across LangGraph, CrewAI, n8n, and the ChatGPT Agents SDK. The gateway emits signed AARM receipts for every decision class across all five authorization outcomes.

Classification

Coverage surface

MCPEndpointSaaSCloudData/DBAPINetwork

Stage

Launched

Type

Commercial

Target audience

Enterprise

Deployment

Self-hostedSaaSHybrid

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 level

Extended (R1–R9)

Conformance review

Platform capabilities

• Pre-execution MCP gateway intercepting tool calls inline before execution
• All five AARM authorization decisions with traceable receipts
• Identity binding via Okta and Google OAuth with deny-on-missing-identity
• Signed Ed25519 receipts with hash-chained context fields for tamper detection
• Inline PII detection and redaction (emails, SSNs, and other entity classes)
• AI Agent Scope Guard with per-agent natural-language scope definitions
• Semantic distance tracking across extended agent sessions (R7)
• Telemetry export to Splunk with documented JSON schema (R8)
• JIT credential issuance integrating HashiCorp Vault, CyberArk, AWS IAM (R9)

Architecture

Operant enforces AARM requirements at the MCP Gateway, a pre-execution interception point between the agent client and downstream MCP servers. Every tool call routes through the gateway, where the request is parsed at the protocol level, sensitive-entity scanners classify parameters, accumulated session context is loaded, the applicable policy is evaluated, and a decision is returned inline before the tool executes. Session context is captured and chained across the full agent loop — a single request ID links the original user prompt, every subsequent tool call, every detection event, and the final agent response. Every decision produces an AARM receipt containing action descriptors, requester context, policy decision, execution outcome, and an Ed25519 signature with key identifier.

Key facts