Task Orchestration Protocol: A Technical Specification for Semantic Task Orchestration in Knowledge-Intensive Multi-Agent Systems

Existing orchestration approaches - from the Model Context Protocol (MCP) and Business Process Management (BPMN) to agent-based frameworks such as LangGraph and CrewAI - each solve partial problems of task coordination. None of them defines what a task _is_ as a semantic object: with formal role assignments, explicit context bindings, atomic process structure, and governed artefact production. This absence is inconsequential for simple automation but becomes a structural liability in knowledge-intensive work, where tasks are heterogeneous, responsibilities are distributed between human and AI actors, and accountability cannot be delegated to a machine. The Task Orchestration Protocol (TOP) closes this gap. TOP defines a task as a three-layer object - Template, Instance, and Contract - separating capability declaration from runtime execution and from immutable audit records. Tasks register with an orchestrator via a four-phase capability protocol, declare prerequisites and context bindings explicitly, produce versionable artefacts with full lineage, and trigger downstream tasks through typed events. Responsibilities are governed by a formal, rule-based RASCI model with one inviolable constraint: Accountable (A) is always a human role. MCP serves as the universal communication substrate for all task relationships - not only tool invocations, but role interactions, context retrieval, approval requests, and notifications. Orchestrators form a federated peer network without a mandatory central master. TOP is transport-agnostic, formally complementary to both MCP and Contractual Task Orchestration (CTO), and designed for deployment in regulated, knowledge-intensive domains such as business engineering, requirements elicitation, and public administration - where decisions must remain explainable, traceable, and defensible long after the work is done.