Zero-Trust AI Security & Governed MCP Gateway for the Enterprise.
Discover shadow AI agents across Cursor and Claude Desktop, eliminate prompt injection attacks, broker Just-In-Time (JIT) credentials, and enforce granular RBAC over all Model Context Protocol (MCP) tool integrations.
↳ Risk Score: 85/100 (Unsanctioned Exfiltration vector)
↳ Action: Quarantined & Enrolled into Governed Gateway
Gain Full Visibility Over Every AI Agent in Your Enterprise
Enterprise developers and business units spawn autonomous AI copilots and MCP tools daily. Gatewright continuously discovers, scores, and governs every shadow agent before data leaves your boundary.
Autonomous Discovery
Automatically detects unmanaged agent runtimes across developer laptops, cloud VMs, and Kubernetes pods—including Cursor IDE, Claude Desktop, Windsurf, and LangChain/CrewAI scripts.
MITRE ATLAS Risk Scoring
Evaluates every connected agent and tool against the MITRE ATLAS matrix, checking for Prompt Injection (AML.T0054), Excessive Agency (AML.T0055), and Unsanctioned Exfiltration.
Attack Path Graph
Maps identity relations from corporate Okta / Entra ID users to AI agents, bound MCP servers, and downstream sensitive databases—identifying blast radius risks instantly.
Scoped Zero-Trust Endpoints for Every Enterprise Tool
Never route all enterprise tools through a single monolithic endpoint. Gatewright provisions isolated, scoped proxy endpoints for each service, enforcing strict least-privilege boundaries.
Privilege Isolation via Scoped Endpoints
Every MCP server runs on a dedicated route (e.g. /mcp/postgres, /mcp/k8s, /mcp/github). Clients pointing to GitHub cannot call Postgres DDL or AWS S3 tools, strictly eliminating cross-server privilege escalation.
❌ Cross-tool invocation blocked (40304)
Just-In-Time (JIT) Credential Broker
AI models and developer workstations never touch database passwords, AWS secret keys, or Kubeconfigs. Gatewright passes opaque tokens like {{GW_VAULT:prod_db_pass}} and injects real secrets dynamically in-memory.
✓ Real credentials never leaked to LLM
Sub-Millisecond Guardrails & DLP
Every tool execution passes through sub-2ms deterministic Open Policy Agent (OPA) checks, semantic guardrails, and data loss prevention (DLP) to redact customer PII, credit cards, and secrets in real-time.
✓ Audited in Trajectory Stream (:8090)
Multi-Cluster Administration with Granular Entitlement Grants
Manage multiple heterogeneous Kubernetes clusters, production databases, and cloud accounts centrally. Bind developers and AI agents to specific backend targets with strict tool-level scoping.
1. Admin Target Infrastructure Connections
Managed at Central Admin Level{{GW_VAULT:k8s_prod_token}}{{GW_VAULT:k8s_staging_kubeconfig}}{{GW_VAULT:prod_db_pass}}{{GW_VAULT:analytics_replica_pass}}2. Access Entitlements & Principal Grants Matrix
User / Team / Agent → Bound Targets| Principal Identity | Scoped MCP Endpoint | Bound Target Cluster | Allowed Tool Scope | Step-Up Approval | Status |
|---|---|---|---|---|---|
| TEAMPlatform SRE Team & CrewAI | /mcp/k8s | AWS EKS Production (US-East) | ALL 10 TOOLS | Required for exec_pod_command | ACTIVE |
| TEAMEngineering Developers | /mcp/k8s-staging | GKE Staging Cluster (EU-West) | READ_ONLY (list_pods, get_pod_logs) | Not Required (Safe Read) | ACTIVE |
| AGENTCursor Lead Copilot (agent-cursor-01) | /mcp/postgres | PostgreSQL Primary (Customer DB) | ALL 10 SQL TOOLS | Required for drop_table & write_query | ACTIVE |
How Gatewright Protects Your Organization
Compare vulnerable shadow agent execution with Gatewright's Zero-Trust guardrail pipeline across common enterprise attack vectors.
Indirect Prompt Injection in Web Content
An attacker embeds a hidden prompt injection in a public documentation page to extract AWS credentials from the agent.
Agent scrapes page → executes injected instructions → dumps AWS_SECRET_ACCESS_KEY to external webhook.
Gatewright Indirect Prompt Injection Scanner intercepts scraped DOM → flags injection payload → suppresses exfiltration tool call.
Governed Connectivity Across Your Entire Stack
Gatewright provides unified Zero-Trust governance, Just-In-Time credential brokering, and real-time DLP for all major cloud runtimes, databases, developer tools, and SaaS platforms.
Cloud & Infrastructure
Governed ConnectorsMulti-cluster operator, namespace isolation, and pod exec guardrails
IAM least-privilege scoping, S3 DLP scans, and JIT STS credential broker
Resource governance and cloud infrastructure proxy controls
Databases & Data Warehouses
Governed ConnectorsQuery-level guardrails, destructive DDL prevention, and in-flight DLP masking
Column-level PII redaction and query cost protection bounds
In-memory cache access control and key-level permission enforcement
Developer & DevOps Platforms
Governed ConnectorsSecret leak prevention in commits, PR review bots, and branch guards
Ticket automation with JQL query sanitization and sprint access controls
Incident triage bots with read-only monitoring and on-call step-up approvals
SaaS, CRM & Business Core
Governed ConnectorsField-level CRM access control and customer PII masking
Interactive human-in-the-loop step-up approval triggers and bot isolation
High-impact transaction guards for refunds, charges, and payouts
Security, Secrets & Identity
Governed ConnectorsDynamic ephemeral secrets resolution and PKI certificate brokering
Corporate identity provider mapping and user-to-agent blast radius tracking
Web Scraping & Intelligence
Governed ConnectorsSSRF private network filtering and DOM prompt injection scanning
Privacy-first web retrieval with indirect prompt injection defense
Built for Regulated Industries & Security Audits
Every prompt, MCP tool execution, and DLP redaction produces an immutable, cryptographically verifiable trajectory log ready for compliance reporting.
SOC 2 Type II
Immutable trajectory audit logging, JIT credential broker with zero secret retention, and granular RBAC for all agent operations.
ISO/IEC 42001
Formal AI security policy definitions, continuous AISPM shadow discovery, and automated mitigation of excessive agency risks.
NIST AI RMF 1.0
Full alignment with the NIST AI Risk Management Framework, tracking risk scores and model blast radius across all infrastructure targets.
EU AI Act & GDPR
Real-time DLP redacts customer PII in-flight and logs step-up human authorization trails for high-privilege tool executions.
Start Free with Open Source. Scale with Enterprise Zero-Trust.
Deploy Gatewright on your own infrastructure with Docker Compose or Helm in minutes, or partner with us for fully-managed enterprise control planes.
Self-Hosted Community Edition
Complete core AI security & MCP gateway suite. Run locally on workstations or deploy to Kubernetes clusters.
/mcp/{slug})Enterprise Zero-Trust Platform
For organizations managing hundreds of AI developers, multiple production clusters, and strict compliance mandates.