Human-Agent Cowork MCP Server

Example Use Cases

• 1

  Propose actions and receive trust-based approval before execution.

• 2

  That require transparent action attribution and audit logging for governance.

• 3

  Escalate tasks to humans with structured context and handle overrides.

Description

This MCP server adds collaboration primitives to AI agents, enabling intent declaration, trust scoring, human handoffs, action attribution, and override handling. It includes a high-performance security proxy called cowork-sentry that enforces action attribution with immutable audit logs, tamper-evident hash chains, and PII sanitization. The system improves AI governance by making agent actions transparent, accountable, and auditable, while allowing trust to be earned or degraded based on overrides.