This repository provides four production-oriented orchestration strategies, each packaged as a separate MCP server so customers can choose the model that best fits their workload.
- Centralized Role Queue
- Best for predictable coordination and straightforward operations.
- Entry point:
agent_orchestrator.mcp_centralized
- Event-Sourced Blackboard
- Best for replay, auditability, and postmortem reconstruction.
- Entry point:
agent_orchestrator.mcp_blackboard
- Hierarchical Planner + Executors
- Best for long-horizon decomposition into DAGs.
- Entry point:
agent_orchestrator.mcp_hierarchical
- Market/Bidding Scheduler
- Best for heterogeneous agents with dynamic assignment.
- Entry point:
agent_orchestrator.mcp_market
PYTHONPATH=src python3 -m agent_orchestrator.mcp_centralized
PYTHONPATH=src python3 -m agent_orchestrator.mcp_blackboard
PYTHONPATH=src python3 -m agent_orchestrator.mcp_hierarchical
PYTHONPATH=src python3 -m agent_orchestrator.mcp_marketPYTHONPATH=src python3 -m agent_orchestrator.server --host 127.0.0.1 --port 8787Detailed implementation guidance for your idea and three alternatives is in:
docs/IMPLEMENTATION_PLAYBOOK.md
That playbook includes:
- deployment steps for each model
- controls for reliability, security, and observability
- model selection guidance
Scenario-specific tests are included for every orchestration strategy and MCP layer:
tests/test_centralized.pytests/test_store.pytests/test_blackboard.pytests/test_hierarchical.pytests/test_market.pytests/test_mcp_servers.py
Run all tests:
PYTHONPATH=src python3 -m unittest discover -s tests -vpyproject.toml exposes executable scripts:
agent-orchestrator-httpagent-orchestrator-mcp-centralizedagent-orchestrator-mcp-blackboardagent-orchestrator-mcp-hierarchicalagent-orchestrator-mcp-market