mcp-agent-orchestrator 0.4.1

MCP server for orchestration, lane management, worker daemons, and turn metrics.

Agent Orchestrator MCP

MCP server for orchestration, lane management, worker daemons, review dispatch, and ACE metrics.

Installation

From PyPI (recommended)

pip install mcp-agent-orchestrator
# or, as an isolated tool:
uv tool install mcp-agent-orchestrator
# or, ad-hoc without installing:
uvx mcp-agent-orchestrator --help

mcp-agent-orchestrator declares mcp-agent-handoff>=0.5.0,<0.6.0 as a required dependency; pip resolves it from PyPI automatically. codex-subagent-bridge remains optional unless you want the local bridge backend.

From the monorepo source tree (development)

From this package root inside agentic-protocol-monorepo:

cd packages/mcp-agent-orchestrator
python -m pip install -e ".[dev]"

When developing both MCP servers in lockstep, install the sibling handoff package as an editable first so the orchestrator picks it up:

pip install -e ../mcp-agent-handoff
pip install -e ".[dev]"

Development

Run package-local commands from the package root:

make lint-orchestrator
make fix-lint-orchestrator
make format-orchestrator
make mypy-orchestrator
make test-orchestrator
make check-orchestrator

The package Makefile keeps codex-subagent-bridge as an optional sibling source path for local bridge-backend development, but it expects mcp-agent-handoff to be installed as a normal package dependency.

Direct commands also work:

PYTHONPATH=src python -m ruff check src tests
PYTHONPATH=src python -m mypy src
PYTHONPATH=src python -m pytest tests -q

Token-Efficient Usage

For bounded reads and compact caller patterns, follow the shared guide in packages/mcp-agent-handoff/docs/guides/token-efficient-usage.md. The orchestrator package reuses that guidance instead of maintaining a separate copy of the same parameter semantics.

Runtime Notes

This package orchestrates work against a target workspace. The workspace you point it at still needs the expected task state and orchestration inputs, such as:

• .task-state/
• lane manifests
• task plans or other orchestration docs the lane logic references

Those assets belong to the workspace being orchestrated, not to the package checkout itself.

Backends

The orchestration layer supports multiple execution backends, including:

• codex-cli
• codex-subagent
• claude-code
• local-model-openai

Some backends are optional and require host-specific tooling to be installed separately.

Source Checkout Usage

For local source execution without installation:

PYTHONPATH=src python -m agent_orchestrator_mcp --help

If you are testing against a sibling codex-subagent-bridge checkout instead of an installed bridge dependency, extend PYTHONPATH with that sibling src directory as needed.