Deterministic Agent Control Protocol

Example Use Cases

• 1

  Bounded and scoped execution of actions to prevent unauthorized operations.

• 2

  Full audit trails with tamper-evident logging of all actions and decisions.

• 3

  Policy-driven governance with human approval gating and session-level budgeting.

Description

This MCP server acts as a control plane that evaluates and enforces policies on AI agent actions, ensuring all operations are bounded within allowed scopes, auditable via a tamper-evident ledger, reversible through compensation plans, and explainable with full reporting. It supports integration with popular AI agents like Cursor, Claude Code, and Codex, and offers multiple integration modes including MCP proxy, shell proxy, HTTP API, and a TypeScript SDK. The server also features policy self-evolution to incrementally relax policies based on user approval, maintaining strict governance while allowing flexibility.