Skip to content

Repository navigation

Security: ONMARTECH/Metabase-AI-MCP-Server

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
5.x.x Yes (Current)
4.x.x Yes (Maintenance)
< 4.0 No

Security Architecture & Defenses (v5.1)

Metabase AI Assistant implements multi-layer defense-in-depth security:

1. AI Provenance Envelopes (_provenance)

All generative SQL, explainers, and dashboard layouts include structured metadata headers and _provenance envelopes warning downstream agents and human reviewers that the content was generated by an AI model.

2. Prompt Injection Defense & Metadata Sandboxing

All database metadata, column comments, and table definitions fetched from Metabase are wrapped in strict [UNTRUSTED_METADATA] boundary delimiters. The system prompt enforces that AI models treat metadata as passive schema data, neutralizing indirect prompt injection payloads.

3. Zero-Leak Enterprise PII Masker (src/utils/pii-masker.js)

Real-time regex and checksum-based sanitization layer inspects query results and data samples, automatically neutralizing:

  • Email addresses ([EMAIL_MASKED])
  • International & Turkish phone numbers
  • Credit card numbers (Luhn validated)
  • National ID numbers (TCKN, SSN)
  • IPv4 / IPv6 addresses
  • API keys, OAuth tokens (ghp_, sk_, xoxb-), and bearer secrets

4. Governance-First Business Memory

  • Zero Silent Learning: Business rules and metric definitions are proposed in PENDING_APPROVAL status and require explicit steward sign-off (semantic_memory_approve).
  • No Hard Deletes: Deprecated rules are safely soft-archived with mandatory reason auditing (semantic_memory_deprecate).

5. Strict Read-Only Gatekeeper

When METABASE_READ_ONLY_MODE=true is set, all mutating tools (DML, DDL, table drops, dashboard deletion, user alterations) are blocked at the MCP router level before execution.

Reporting a Vulnerability

We take security seriously. If you discover a security vulnerability within Metabase AI Assistant, please follow the responsible disclosure process below:

How to Report

  1. Email: Send a detailed report to security@onmartech.com or contact@onmartech.com
  2. Subject Line: Use [SECURITY] Metabase AI Assistant - Vulnerability Report
  3. GitHub Private Advisory: You can also submit via GitHub Security Advisories.

Response Timeline

  • Initial Response: Within 24-48 hours
  • Status Update: Within 5 business days
  • Resolution & Patch Target: Within 14 days for critical vulnerabilities

Contact

Copyright 2024-2026 ONMARTECH LLC. All rights reserved.

There aren't any published security advisories