Security Is Not
an Afterthought
MCP360 connects AI agents to your ad accounts, CRM, and financial tools. Every layer is built with security as the primary design constraint.
Defense in Depth
TLS 1.3
All data encrypted in transit with TLS 1.3. HTTP Strict Transport Security (HSTS) enforced on every connection. No fallback to older protocols.
AES-256-GCM
OAuth tokens and sensitive credentials are encrypted at rest using AES-256-GCM authenticated encryption. Tokens are decrypted only in-memory during tool execution.
Per-Org Encryption Keys
Each organization gets a unique encryption key derived via HKDF. A compromise of one organization cannot expose another. Keys are never stored alongside ciphertext.
OAuth 2.0
All platform connections use standard OAuth 2.0 authorization code flow. We request minimum necessary scopes and handle automatic token refresh.
Rate Limiting
Per-organization, per-API-key, and per-tool rate limits prevent abuse and runaway automation. Configurable thresholds with automatic backoff.
Input Validation
Every MCP tool endpoint validates inputs against strict Zod schemas. Malformed requests, unexpected fields, and out-of-range values are rejected before execution.
Multi-Tenant Isolation
PostgreSQL Row-Level Security (RLS) at the database layer, Prisma middleware at the ORM layer, and service-layer ownership checks provide defense in depth.
Complete Audit Trail
Every MCP tool call is logged with timestamp, user, organization, tool name, platform, and result. Audit logs are immutable and retained for 12 months.
MCP Threat Model
MCP introduces unique attack surfaces. We address each one explicitly.
Tool Poisoning
Strict Zod schema validation on every tool input and output. Tool definitions are immutable and cannot be modified by user prompts. All responses use structured data, never raw instruction text.
Prompt Injection
MCP tool responses return structured JSON data, never instructions that could alter AI behavior. User-supplied strings are never interpolated into tool definitions or system prompts.
Secrets Redaction
OAuth tokens, API keys, and client secrets are never included in MCP tool responses. Internal IDs and infrastructure details are stripped before data reaches the AI client.
Budget Safety
Write operations that modify budgets enforce configurable maximum limits. Changes above threshold require explicit approval via the dashboard. Emergency kill switches halt all write operations instantly.
Compliance
SOC 2 Type II
We are actively pursuing SOC 2 Type II certification. Our infrastructure, access controls, encryption, and audit logging are designed to meet SOC 2 trust service criteria from day one.
GDPR
Data minimization, right to deletion, data export, and consent management are built into the platform. We process only the data necessary to fulfill your MCP tool requests and retain logs for a defined period.
Questions About Security?
We welcome responsible disclosure reports and security inquiries. Reach out any time.