Worker Class Protocol
The open standard for governing AI worker dispatch.
The governance gap
87% of AI agent systems lack safety cards. MCP, A2A, ACP, and AGNTCY define how agents communicate. None define whether a worker should be trusted to execute — under what controls, with what blast radius, and with what evidence.
WCP fills that gap. It is the governance layer between a capability request and its execution.
Where WCP fits
H2A → Human-to-Agent (Discord, CLI, API) AGP → Agent Governance (future) WCP → Worker Dispatch ← THIS SPEC MCP → Model Context (Anthropic) A2A → Agent-to-Agent (Google / Linux Foundation)
WCP governs workers. MCP and A2A transport requests. WCP does not define a transport.
What WCP defines
Worker types, species IDs, risk tiers, and registry records. Every worker declares its capabilities, blast radius, and required controls before it can be dispatched.
Policy-aware dispatch decisions. The Hall matches capability requests to enrolled workers, applies risk policy, and produces a RouteDecision with full audit trail.
Required and suggested constraints on worker execution. A required control not present blocks dispatch. Every control application is recorded.
Every RouteDecision is logged. Correlation IDs link requests to decisions. Attestation hashes prove registry records haven't been tampered with.
Status
Submit comments via GitHub Issues. WCP is open — implementations in Python, TypeScript, and Go are available at pyhall.dev.