TORK MCP Server

Governance tools for AI agents via Model Context Protocol (MCP).

Part of the TORK Network - AI governance infrastructure for the agentic era.

Installation

npm install @torknetwork/mcp-server

Features

The TORK MCP Server provides 21 governance tools across four categories:

Core Governance Tools

| Tool | Description | |------|-------------| | tork_scan_pii | Scan text for PII (SSN, credit cards, emails, phones, IP addresses, API keys) and optionally redact | | tork_governed_passthrough | Execute tool calls with full governance checks (validation + audit) | | tork_check_policy | Check if an action is allowed by policy without executing it | | tork_audit_stats | Get audit statistics for governance events | | tork_audit_logs | Query audit logs for governance events |

Tool Call Governance

| Tool | Description | |------|-------------| | tork_register_tool | Register a tool with governance policy (allowed agents, targets, blocked flags, rate limits, HITL requirements) | | tork_validate_tool_call | Validate a tool call BEFORE execution - checks registration, agent authorization, target restrictions, blocked flags, and rate limits | | tork_tool_call_stats | Get statistics on tool calls including allowed/blocked counts and rate limit hits |

HITL (Human-in-the-Loop) Enforcement

| Tool | Description | |------|-------------| | tork_request_approval | Submit an action for human approval with velocity tracking for slicing attack detection | | tork_approve_action | Human approves a pending action (with fatigue detection and cool-down enforcement) | | tork_reject_action | Human rejects a pending action (rejections indicate vigilance, no cool-down penalty) | | tork_check_approval | Check the status of an approval request | | tork_velocity_alert | Check for slicing attack patterns - many small requests aggregating to significant impact |

Cloud API Tools (v1.1.0+)

These tools connect to the live Tork Cloud API for centralized governance management. Requires TORK_API_KEY environment variable.

| Tool | Description | |------|-------------| | tork_dashboard | Get dashboard statistics (total tools, agents, policies, pending approvals) | | tork_list_tools | List all registered tools from Tork Cloud | | tork_list_policies | List all governance policies | | tork_get_policy | Get a specific policy by name | | tork_list_approvals | List pending HITL approval requests | | tork_approve | Approve a pending HITL request | | tork_deny | Deny a pending HITL request with reason | | tork_cloud_audit_logs | Query audit logs with filtering | | tork_create_webhook | Create a webhook for HITL notifications (v1.2.0+) | | tork_list_webhooks | List all configured webhooks (v1.2.0+) | | tork_delete_webhook | Delete a webhook by ID (v1.2.0+) | | tork_governance_check | Real-time policy validation against cloud (v1.2.0+) |

Getting an API Key

1. Sign up at tork.network/signup
2. Create an account with your email and password
3. Your API key will be displayed after signup (save it securely!)
4. Set the environment variable: export TORK_API_KEY=tork_live_xxxxx

Or via API:

curl -X POST https://tork.network/api/v1/auth/signup \
  -H "Content-Type: application/json" \
  -d '{"email":"[email protected]","password":"SecurePass123!","organization_name":"My Company"}'

Usage

With Claude Desktop

Add to your Claude Desktop configuration (~/Library/Application Support/Claude/claude_desktop_config.json):

Basic (local mode only):

{
  "mcpServers": {
    "tork": {
      "command": "npx",
      "args": ["@torknetwork/mcp-server"]
    }
  }
}

With Cloud API (recommended):

{
  "mcpServers": {
    "tork": {
      "command": "npx",
      "args": ["@torknetwork/mcp-server"],
      "env": {
        "TORK_API_KEY": "tork_live_your_api_key_here"
      }
    }
  }
}

Example Prompts for Claude Desktop

Once configured, try these prompts with Claude:

Dashboard & Overview:

• "Show me the Tork governance dashboard"
• "List all registered tools in my Tork account"
• "What governance policies do I have configured?"

Approvals & HITL:

• "Show me pending approval requests"
• "Approve request tork_abc123 as reviewer [email protected]"
• "Deny request tork_xyz789 because it targets production without authorization"

Audit & Compliance:

• "Show me recent audit logs"
• "Query audit logs for agent-prod-1 from the last 24 hours"

PII Scanning:

• "Scan this text for PII: Contact [email protected] at 555-1234"
• "Check if this message contains sensitive data"

Programmatic Usage

import { spawn } from 'child_process';

const server = spawn('npx', ['@torknetwork/mcp-server'], {
  stdio: ['pipe', 'pipe', 'inherit']
});

// Send MCP requests via stdin, receive responses via stdout

Example: Register a Tool with Governance

{
  "tool": "tork_register_tool",
  "arguments": {
    "tool_name": "database_write",
    "allowed_agents": ["agent-prod-1", "agent-prod-2"],
    "allowed_targets": ["staging", "dev"],
    "blocked_flags": ["--drop-table", "--truncate"],
    "max_calls_per_minute": 10,
    "requires_hitl": true
  }
}

Example: Validate Before Execution

{
  "tool": "tork_validate_tool_call",
  "arguments": {
    "agent_id": "agent-prod-1",
    "tool_name": "database_write",
    "target_environment": "staging",
    "parameters": {
      "query": "INSERT INTO users (name) VALUES ('test')"
    }
  }
}

Security Features

• PII Detection: Automatically scans for and redacts sensitive data
• Blocked Flags: Prevents dangerous flags like --force, --no-confirm, --delete-all
• Rate Limiting: Configurable per-tool rate limits
• Agent Authorization: Control which agents can call which tools
• Target Restrictions: Prevent production access from unauthorized contexts
• HITL Enforcement: Require human approval for sensitive operations
• Fatigue Detection: Lock reviewers who approve too many requests too quickly
• Slicing Attack Detection: Detect patterns of many small requests aggregating to significant impact

Configuration

HITL enforcement uses these defaults (configurable in code):

const HITL_CONFIG = {
  max_approvals_per_minute: 5,      // Triggers cool-down after this many approvals
  cool_down_minutes: 15,            // Duration of reviewer lock-out
  velocity_window_minutes: 10,      // Window for detecting slicing attacks
  velocity_threshold_count: 10,     // Max requests before velocity alert
  velocity_threshold_amount: 10000, // Max cumulative amount before velocity alert
  approval_expiry_minutes: 30,      // Approvals expire if not reviewed
};

License

MIT - see LICENSE

Links

• TORK Network
• GitHub Repository
• Report Issues