MCP Tool Policies & Enterprise Governance

Enforce granular security guardrails, allowlists, and execution approval workflows on tools exposed via Model Context Protocol (MCP) servers.

• Portal Page: /projects/[projectId]/mcp-policies

graph LR
    ToolCall["Agent Tool Call (drop_table)"] --> PolicyEngine["Naagmani MCP Policy Engine"]
    PolicyEngine --> Check{"Policy Decision"}
    Check -->|Allowed| Exec["Execute Tool"]
    Check -->|Blocked| Block["Reject (HTTP 403 Security Policy)"]
    Check -->|Requires Approval| Notify["Pause & Notify Human Reviewer"]

Policy Types #

1. Tool Allowlists & Blocklists: Explicitly allow or prohibit specific tools (e.g. allow read_file, block delete_file).
2. Parameter Constraints: Restrict tool parameters (e.g., limit SQL queries to SELECT statements only).
3. Human-in-the-Loop Approvals: Pause destructive tool executions until an administrator approves the action.
4. Rate Limiting: Cap the number of times an agent can invoke a specific tool per minute.

Next Steps #

• Configure content guardrails: AI Guardrails & Governance
• View security audit trail: Security Audit Logs