MCP Policies

Field	Value
Document ID	ASCEND-MCP-002
Version	1.0.0
Last Updated	December 19, 2025
Author	Ascend Engineering Team
Publisher	OW-KAI Technologies Inc.
Classification	Enterprise Client Documentation
Compliance	SOC 2 CC6.1/CC6.2, PCI-DSS 7.1/8.3, HIPAA 164.312, NIST 800-53 AC-2/SI-4

Reading Time: 8 minutes | Skill Level: Intermediate

Overview

MCP (Model Context Protocol) policies define how ASCEND governs interactions between AI models and MCP servers. Configure trust levels, tool restrictions, and approval requirements.

Policy Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    MCP GOVERNANCE LAYER                          │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌─────────────┐    ┌─────────────┐    ┌─────────────┐         │
│  │   Server    │    │    Tool     │    │   Action    │         │
│  │  Policies   │    │  Policies   │    │  Policies   │         │
│  └──────┬──────┘    └──────┬──────┘    └──────┬──────┘         │
│         │                  │                  │                 │
│         └──────────────────┼──────────────────┘                 │
│                            │                                    │
│                            ▼                                    │
│                   ┌─────────────────┐                          │
│                   │ Unified Policy  │                          │
│                   │    Engine       │                          │
│                   └────────┬────────┘                          │
│                            │                                    │
│              ┌─────────────┼─────────────┐                     │
│              ▼             ▼             ▼                     │
│         ┌────────┐   ┌────────┐   ┌────────┐                  │
│         │ Allow  │   │ Require│   │  Deny  │                  │
│         │        │   │Approval│   │        │                  │
│         └────────┘   └────────┘   └────────┘                  │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Server Trust Levels

Trust Level Definitions

Level	Description	Default Behavior
trusted	Fully vetted, internal servers	Auto-approve most actions
restricted	Partially vetted, limited scope	Require approval for sensitive
sandbox	Untested or external servers	Require approval for all

Configure Server Trust

curl -X PUT "https://pilot.owkai.app/api/mcp/servers/{server_id}/trust" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "trust_level": "restricted",
    "reason": "Third-party integration, limited to read operations",
    "allowed_tools": ["search", "read_file"],
    "denied_tools": ["write_file", "execute", "delete"],
    "max_risk_score": 50
  }'

Tool Policies

Create Tool Policy

curl -X POST "https://pilot.owkai.app/api/mcp/policies/tools" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "name": "File Write Approval",
    "tool_pattern": "write_*",
    "servers": ["*"],
    "conditions": {
      "parameters.path": {"$regex": "^/sensitive/"}
    },
    "action": "REQUIRE_APPROVAL",
    "approval_level": 2,
    "risk_modifier": 30
  }'

Tool Pattern Matching

Pattern	Matches
*	All tools
read_*	read_file, read_database, etc.
*_database	read_database, write_database
execute	Exact match only

List Tool Policies

curl "https://pilot.owkai.app/api/mcp/policies/tools" \
  -H "Authorization: Bearer <jwt_token>"

Response:

{
  "policies": [
    {
      "policy_id": "pol_mcp_001",
      "name": "File Write Approval",
      "tool_pattern": "write_*",
      "action": "REQUIRE_APPROVAL",
      "is_active": true,
      "priority": 100
    },
    {
      "policy_id": "pol_mcp_002",
      "name": "Block Execute on Sandbox",
      "tool_pattern": "execute",
      "servers": ["sandbox_*"],
      "action": "DENY",
      "is_active": true,
      "priority": 200
    }
  ]
}

Server Policies

Create Server Policy

curl -X POST "https://pilot.owkai.app/api/mcp/policies/servers" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "name": "External Server Restrictions",
    "server_pattern": "external_*",
    "default_action": "REQUIRE_APPROVAL",
    "max_requests_per_minute": 10,
    "allowed_ip_ranges": ["10.0.0.0/8"],
    "require_mTLS": true,
    "audit_all_requests": true
  }'

Server Access Control

curl -X PUT "https://pilot.owkai.app/api/mcp/servers/{server_id}/access" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "allowed_agents": ["agent_001", "agent_002"],
    "denied_agents": ["untrusted_agent"],
    "require_agent_registration": true,
    "ip_whitelist": ["192.168.1.0/24"]
  }'

Unified Governance

Agent + MCP Unified Policy

curl -X POST "https://pilot.owkai.app/api/governance/unified" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "name": "Sensitive Data Access",
    "applies_to": ["agents", "mcp_servers"],
    "conditions": {
      "$or": [
        {"data_category": "pii"},
        {"data_category": "financial"},
        {"data_category": "health"}
      ]
    },
    "action": "REQUIRE_APPROVAL",
    "approval_level": 3,
    "audit_retention_days": 2555
  }'

Policy Evaluation Order

1. Deny policies (highest priority)
2. Server-specific policies
3. Tool-specific policies
4. Trust level defaults
5. Global defaults

Risk Modifiers

Configure MCP Risk Scoring

curl -X PUT "https://pilot.owkai.app/api/mcp/risk-config" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "tool_modifiers": {
      "execute": 50,
      "write_*": 30,
      "delete_*": 40,
      "read_*": 5,
      "search": 0
    },
    "server_modifiers": {
      "sandbox": 30,
      "restricted": 15,
      "trusted": 0
    },
    "parameter_modifiers": {
      "contains_pii": 25,
      "external_destination": 20,
      "bulk_operation": 15
    }
  }'

Policy Templates

Apply Security Template

curl -X POST "https://pilot.owkai.app/api/mcp/policies/templates/security-strict/apply" \
  -H "Authorization: Bearer <admin_jwt>"

Security-Strict Template:

• All write operations require approval
• Execute tools blocked by default
• External server requests denied
• Full audit logging

Available Templates

Template	Description
security-strict	Maximum restrictions
security-standard	Balanced security
development	Relaxed for dev/test
compliance-hipaa	HIPAA-compliant settings
compliance-pci	PCI-DSS compliant settings

Conditional Policies

Time-Based Restrictions

curl -X POST "https://pilot.owkai.app/api/mcp/policies" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "name": "After Hours Restrictions",
    "conditions": {
      "time_of_day": {"$gte": "18:00", "$lte": "08:00"},
      "day_of_week": {"$nin": ["saturday", "sunday"]}
    },
    "action": "REQUIRE_APPROVAL",
    "approval_level": 3
  }'

Parameter-Based Policies

curl -X POST "https://pilot.owkai.app/api/mcp/policies" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "name": "Large Data Export Approval",
    "tool_pattern": "export_*",
    "conditions": {
      "parameters.row_count": {"$gte": 10000}
    },
    "action": "REQUIRE_APPROVAL",
    "approval_level": 2
  }'

Policy Testing

Test Policy Against Request

curl -X POST "https://pilot.owkai.app/api/mcp/policies/test" \
  -H "Authorization: Bearer <admin_jwt>" \
  -d '{
    "server_id": "external_api_server",
    "tool_name": "write_file",
    "parameters": {
      "path": "/sensitive/data.json",
      "content": "test"
    },
    "agent_id": "test_agent"
  }'

Response:

{
  "decision": "REQUIRE_APPROVAL",
  "matched_policies": [
    {
      "policy_id": "pol_mcp_001",
      "name": "File Write Approval",
      "priority": 100
    }
  ],
  "risk_score": 65,
  "risk_factors": [
    {"name": "write_operation", "score": 30},
    {"name": "sensitive_path", "score": 25},
    {"name": "restricted_server", "score": 10}
  ]
}

Audit & Monitoring

MCP Activity Logs

curl "https://pilot.owkai.app/api/mcp/audit?server_id=external_api&days=7" \
  -H "Authorization: Bearer <admin_jwt>"

Response:

{
  "logs": [
    {
      "timestamp": "2025-12-15T10:30:00Z",
      "server_id": "external_api",
      "tool_name": "search",
      "agent_id": "research_agent",
      "decision": "approved",
      "risk_score": 15,
      "response_time_ms": 45
    }
  ],
  "statistics": {
    "total_requests": 150,
    "approved": 145,
    "denied": 5,
    "avg_risk_score": 22
  }
}

Best Practices

1. Start Restrictive

Begin with sandbox trust level and restrictive policies, then relax as confidence grows.

2. Use Specific Patterns

Prefer specific tool patterns over wildcards:

# Better
"tool_pattern": "write_database"

# Less precise
"tool_pattern": "write_*"

3. Layer Policies

Combine server, tool, and parameter policies for defense in depth.

4. Monitor and Adjust

Review audit logs regularly and adjust policies based on actual usage patterns.

Next Steps

• MCP Overview - MCP governance basics
• Server Registration - Register MCP servers
• Smart Rules - General policy configuration

---

Document Version: 1.0.0 | Last Updated: December 2025