OSVauco/docs/osvx/mcp/TOOL_POLICY.md

OSVx MCP Tool Policy

This document defines the security classification and exposure tiers for all MCP tools.

Tier 1: Safe Reads

• Description: Public, cacheable, low-risk data.
• Tools: list_emma_models

Tier 2: Scoped Reads

• Description: Requires authentication; reads sensitive but non-modifying data.
• Tools: get_billing_history, get_file, get_emails

Tier 3: Compute Tools

• Description: Requires authentication; triggers non-state-changing computations.
• Tools: run_jason, run_emma

Tier 4: Confirmed Actions

• Description: State-changing actions that should require user confirmation (HITL).
• Tools: create_issue, set_billing_budget

Tier 5: Restricted Admin

• Description: High-risk administrative tools that must be protected by strict, server-side authorization.
• Tools: run_terminal, trigger_build, delete_email_alias