MCP Gateway Security

MCP Gateway Security Skill

🔴 AI FIRST Quality Principle

Apply the AI FIRST principle: never accept first-pass quality. Minimum 2 iterations. Read all output, improve every section. No shortcuts.

Purpose

This skill provides security patterns and best practices for securing MCP (Model Context Protocol) gateway communications in the Riksdagsmonitor platform. It covers token management, request validation, audit logging, and threat mitigation for AI-assisted development workflows.

When to Use This Skill

Apply this skill when:

• ✅ Hardening MCP gateway configurations
• ✅ Managing tokens and credentials for MCP servers
• ✅ Implementing request validation for MCP tool calls
• ✅ Setting up audit logging for MCP communications
• ✅ Conducting security reviews of MCP configurations
• ✅ Responding to security incidents involving MCP
• ✅ Assessing risks of new MCP server integrations

Do NOT use for:

• ❌ Initial MCP setup (use mcp-gateway-configuration)
• ❌ Application-level security (use secure-code-review)
• ❌ Infrastructure security (use security-architecture-validation)

Threat Model for MCP

Attack Surface

┌──────────────────────────────────────────────┐
│                 Threat Vectors                │
├──────────────────────────────────────────────┤
│                                              │
│  ┌──────────┐    ┌──────────┐    ┌────────┐ │
│  │ Token     │    │ Prompt   │    │ Supply │ │
│  │ Theft     │    │ Injection│    │ Chain  │ │
│  └────┬─────┘    └────┬─────┘    └───┬────┘ │
│       │               │              │       │
│       ▼               ▼              ▼       │
│  ┌─────────────────────────────────────────┐ │
│  │         MCP Gateway                      │ │
│  └────┬────────────┬────────────┬──────────┘ │
│       │            │            │            │
│  ┌────▼────┐  ┌────▼────┐  ┌───▼─────┐     │
│  │ Unauth  │  │ Data    │  │ Lateral │     │
│  │ Access  │  │ Exfil   │  │ Movement│     │
│  └─────────┘  └─────────┘  └─────────┘     │
└──────────────────────────────────────────────┘

Threat Matrix

Token Management

Token Security Requirements

Secure Token Configuration

{
  "mcpServers": {
    "github": {
      "type": "stdio",
      "command": "github-mcp-server",
      "env": {
        "GITHUB_TOKEN": "${GITHUB_TOKEN}"
      }
    }
  }
}

Token Handling Rules:

✅ DO: Use environment variable references (${VAR_NAME} in MCP JSON config files; $VAR_NAME in shell scripts; ${{ secrets.VAR_NAME }} in GitHub Actions)
✅ DO: Use GitHub Actions secrets for CI/CD tokens
✅ DO: Rotate tokens at least quarterly
✅ DO: Use fine-grained PATs with minimal scopes
✅ DO: Revoke tokens immediately when compromised

❌ DON'T: Hardcode tokens in configuration files
❌ DON'T: Commit tokens to version control
❌ DON'T: Share tokens between environments
❌ DON'T: Use classic PATs with broad scopes
❌ DON'T: Log token values in any log output

GitHub Token Scopes (Principle of Least Privilege)

Request Validation

Input Validation for MCP Tools

File Operations:

Validation Rules:
1. Path must be within allowed directories
2. Path must not contain traversal sequences (../)
3. File extension must be in allowed list
4. File size must not exceed limits
5. Content must not contain known malicious patterns

Code Operations:

Validation Rules:
1. Branch names must match allowed pattern
2. Commit messages must not contain secrets
3. File content must pass security scanning
4. PR descriptions must not leak sensitive data

Output Sanitization

Before Returning MCP Tool Output:
1. Strip any credential-like patterns
2. Remove internal IP addresses/hostnames
3. Truncate excessively large outputs
4. Validate JSON/structured output format
5. Log sanitization actions for audit

Dangerous Tool Patterns

Audit Logging

What to Log

What NOT to Log

❌ Token values or API keys
❌ File contents containing secrets
❌ User passwords or credentials
❌ Full request/response bodies with PII
❌ Internal network topology details

Audit Log Format

{
  "timestamp": "2024-01-15T10:30:00Z",
  "event": "mcp.tool.invocation",
  "server": "github",
  "tool": "create_pull_request",
  "parameters": {
    "owner": "Hack23",
    "repo": "cia",
    "title": "[REDACTED]"
  },
  "result": "success",
  "duration_ms": 1250,
  "user": "copilot-agent"
}

Security Configuration Checklist

Pre-Deployment

□ All tokens use environment variables (never hardcoded)
□ Token scopes follow least privilege
□ Filesystem access restricted to project directory only
□ MCP package versions pinned to specific releases
□ Configuration file committed (without secrets)
□ No sensitive data in MCP server arguments
□ SSE server URLs use HTTPS only
□ Certificate validation enabled for remote servers

Periodic Review (Monthly)

□ Review token scopes — remove unnecessary permissions
□ Rotate tokens per schedule
□ Check for new MCP package versions and CVEs
□ Review audit logs for anomalies
□ Verify directory restrictions still appropriate
□ Test authentication failure handling
□ Review and update threat model
□ Check for deprecated MCP server versions

Incident Response

MCP Security Incident
    │
    ├─→ Token Compromise
    │   ├─→ Revoke token immediately
    │   ├─→ Rotate all related tokens
    │   ├─→ Review audit logs for unauthorized access
    │   └─→ Update token storage mechanism
    │
    ├─→ Unauthorized File Access
    │   ├─→ Review filesystem server configuration
    │   ├─→ Check for directory traversal attempts
    │   ├─→ Restrict filesystem paths
    │   └─→ Review accessed files for data exposure
    │
    ├─→ Supply Chain Attack
    │   ├─→ Pin to known-good version
    │   ├─→ Verify package integrity
    │   ├─→ Check for malicious tool behavior
    │   └─→ Report to MCP package maintainers
    │
    └─→ Prompt Injection
        ├─→ Review tool invocation logs
        ├─→ Identify injected content
        ├─→ Assess data exposure
        └─→ Update input validation rules

OWASP Agentic Security Alignment

ISMS Alignment

References

• Model Context Protocol Security
• OWASP Agentic AI Security
• GitHub Fine-grained PATs
• Hack23 ISMS Security Policy