# MCP Task Orchestrator

An MCP server that gives AI coding agents a persistent, server-enforced work item graph with quality gates, dependency tracking and actor attribution, instead of relying on the model to follow prompted rules.

Task Orchestrator is an MCP server built for multi-agent or multi-session coding workflows, such as an orchestrator dispatching sub-agents or a fresh session picking up earlier work. It enforces workflow discipline in the server rather than the prompt: if a required design note is missing, advancing a work item returns an error naming what is missing, and a transition is blocked if an upstream dependency is not complete, with every transition and note recording who made it.

It ships as a Docker image, works with any MCP client, and has an optional Claude Code plugin adding skills and hooks. Rules for what documentation must exist before work starts are defined in YAML configuration rather than hardcoded, and a new session can recover full project state in one call instead of replaying a conversation. Without a config file, every tool works in a schema-free mode with no gates.

Website: https://github.com/jpicklyk/task-orchestrator