guardrails-mcp-server

[!License: MIT](https://opensource.org/licenses/MIT) [!MCP](https://modelcontextprotocol.io) [!Node.js](https://nodejs.org)

MCP server for AI agent security guardrails. Provides input validation, prompt injection detection, PII redaction, output filtering, policy enforcement, rate limiting, and comprehensive audit logging.

Tools

Tool	Description
validate_input	Validate and sanitize incoming requests through all guardrail checks
filter_output	Filter and redact sensitive data (PII, secrets, credentials) from responses
check_policy	Evaluate a request against security policies (RBAC, resource access, quotas)
get_audit_logs	Query the audit log with filtering by type, user, time range
get_stats	Get engine statistics including active users, block rate, request counts
update_config	Update guardrail configuration at runtime

Security Features

• Prompt Injection Detection -- 12 regex patterns for jailbreak, DAN mode, system prompt override
• PII Detection and Redaction -- SSN, credit card, email, phone, IP, API keys, AWS keys, JWT, passwords, private keys, connection strings
• Malicious Code Blocking -- eval, exec, subprocess, child_process, shell injection
• Policy Engine -- Block sensitive paths, dangerous tools, unauthenticated destructive ops, URL allowlist, maintenance windows
• Rate Limiting -- Per-user sliding window (configurable requests/minute)
• Audit Logging -- Timestamped events with metrics, log rotation, external handler support

Install

npm install

Configuration

{
  "mcpServers": {
    "guardrails": {
      "type": "stdio",
      "command": "node",
      "args": ["/path/to/guardrails-mcp-server/index.js"]
    }
  }
}

Project Structure

src/engine/GuardrailsEngine.js    # Core orchestration
src/validators/InputValidator.js   # Prompt injection and PII detection
src/filters/OutputFilter.js        # Redaction and harmful content blocking
src/policies/PolicyEngine.js       # RBAC, quotas, maintenance windows
src/audit/AuditLogger.js           # Event logging and metrics

License

MIT