agent-policy-gateway-mcp-js

MCP server for AI agent policy enforcement. Provides PII detection, guardrails, GDPR/EU AI Act compliance checking, audit logging, and an emergency kill switch.

• PII Detection — Scan text for emails, phone numbers, SSNs, credit cards, IBANs with automatic redaction
• Guardrails — Check agent actions against configurable policies (spend limits, domain whitelists, blocked actions)
• Audit Logging — Append-only JSONL audit trail per agent in ~/.agent-audit-log/
• GDPR Compliance — Check data processing actions against DSGVO requirements
• EU AI Act Compliance — Verify AI operations against EU AI Act risk categories
• Emergency Kill Switch — Instantly block an agent with critical-level audit entry

npx @aiagentkarl/agent-policy-gateway-mcp

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "policy-gateway": {
      "command": "npx",
      "args": ["-y", "@aiagentkarl/agent-policy-gateway-mcp"]
    }
  }
}

Scan text for personally identifiable information using regex patterns.

Parameters:

• text (string) — Text to scan

Returns: Found PII types, count, masked examples, redacted text, risk level.

Check an agent action against security policies.

Parameters:

• action (string) — Action to check (e.g. "browse_url", "make_purchase")
• context (object) — Action context with optional agent_id, target_url, amount_usd, details

Returns: allow/deny/warn decision with violations and recommendations.

Write an entry to the append-only audit log.

Parameters:

• agent_id (string) — Unique agent identifier
• action (string) — Action performed
• details (string, optional) — Additional details

Retrieve audit log entries for an agent.

Parameters:

• agent_id (string) — Agent identifier
• limit (number, opt