Any system that records every decision accumulates dead tasks, and this one archives work older than 24 hours, validates the task hierarchy and reports health rather than waiting for the store to become unusable. The specialist roles are worth customising early — the built-in set is generic, and a project-specific role file is where the orchestration starts producing output shaped like your codebase.
An MCP server that turns one large request into an orchestrated sequence of specialist roles, then documents what each one did. It detects your project structure and saves artifacts where they belong, keeps the decisions and implementations from previous sessions so context is not lost, and includes a maintenance pass that archives stale work and keeps the store from bloating.
- A session started against a working directory, then a task broken into a plan — `orchestrator_initialize_session`, `orchestrator_plan_task`
- Each subtask executed with a specialist role's context and completed with its artifacts attached, then combined into a result — `orchestrator_execute_task`, `orchestrator_complete_task`, `orchestrator_synthesize_results`
- Progress readable at any point — `orchestrator_get_status`
- Automated upkeep: stale tasks over 24 hours detected and archived, database bloat prevented, task hierarchies validated, handovers prepared and system health reported — `orchestrator_maintenance_coordinator`
- Built-in specialist roles — architect, implementer, tester, reviewer, documenter — plus project-specific ones you define in `.task_orchestrator/roles/project_roles.yaml`, created automatically when you start a session in a directory
- A template system of 13 tools for reusable task templates
- Workspace awareness: project structure detected so artifacts land in the right place rather than the current directory
No account and no key. Python 3.8+ and `pip install mcp-task-orchestrator`, then add it to your MCP client's configuration; the published project is `mcp-task-orchestrator` 2.0.0 with `mcp-task-orchestrator` and `mcp-task-orchestrator-cli` entry points. The universal installer handles client configuration for Claude Desktop, Cursor IDE, Windsurf and VS Code with the Cline extension. Custom specialist roles are a YAML file in `.task_orchestrator/roles/` inside the project, written the first time you start an orchestration session there.
One command — pip install mcp-task-orchestrator