Maestro
Multi-agent orchestration for Claude Code. Coordinate up to 5 specialised crews in parallel (2 on the free community tier), against a shared workspace, with quality and security gates between roles.
This is the reference documentation for Maestro (CLI: mso). For positioning, pricing, and trust, see the Maestro site.
Heritage note: Maestro was previously named Admiralty. The legacy
adm/admiraltyCLI entry points andADMIRALTY_*environment variables were removed in v4.0 — usemsoandMAESTRO_*exclusively.
Where to start
| Reading path | Best for |
|---|---|
| Installation and configuration | New users: install, licence, mso setup, and the Hub |
| CLI reference | Day-to-day operators |
| Loop engineering vs. graph engineering | Understanding how Maestro structures work |
| Division packs | Tailoring a workspace to your team's discipline |
| Maestro site | Positioning, pricing, and trust |
Installing, upgrading, uninstalling
pip install maestro-fleet
mso version
Maestro runs without a licence at the community tier — permanent and free, not a trial. Activate a Pro, Team or Enterprise licence with mso licence activate <KEY>; mso licence status shows the tier you are running at.
That installs the CLI. To get to a working setup (licence, workspace, the mso setup wizard, and the Hub), follow Installation and configuration.
Every install ships its full documentation set on disk — not just this site. After installing, run mso docs to list every shipped topic, or mso docs <topic> to read one offline (e.g. mso docs install-upgrade-uninstall for the package-rename, scope-shadowing, and uninstall reference; mso docs troubleshooting for numbered error codes; mso docs licence for the exact tier limits).
What's in the box
Features marked with a tier are checked against your licence at the point of use; everything else is available on every tier, including community.
- 7 specialised roles — Planner (PLN), Builder (BLD), Tester (TST), Docs writer (DOC), Security reviewer (SEC), Releaser (REL), Lead (LEAD) — each bound to a model appropriate to its work. Community tier dispatches Planner, Builder and Tester crews; Docs writer, Security reviewer and Releaser crews require Pro, Team or Enterprise. In the nautical persona these roles are known as Navigator, Shipwright, Bosun, Scribe, Lookout, Coxswain, and Quartermaster.
- Voyages and orders — a voyage groups related orders (features, bugs, investigations, security and docs work); one voyage = one branch = one PR. The dispatcher creates the voyage branch when the first crew is dispatched. Community tier is limited to 1 registered project; Pro and above are unlimited.
- Quality gates — per-order acceptance criteria, Tester validation, an optional local verify command that holds a broken build before it is pushed, and optional human review gates between roles. The Releaser (Pro+, as above) reviews CI and review threads and hands you a
READY/NOT READYrecommendation — it never merges, tags, or publishes. - Recovery built in — an activity watchdog for hung crews, salvage branches that preserve a killed crew's in-flight work, and
mso order retryfor stranded orders. - Fleet self-check —
mso doctorruns a read-only health check across dispatcher, crews, orders and voyages, even with the dispatcher stopped. The resident LEAD agent (mso lead start), a background daemon that patrols the fleet and performs routine housekeeping, requires Pro, Team or Enterprise. - Maestro Hub — a multi-workspace dashboard served locally by
mso hub start(no Docker or Node required). Watch the fleet, create and edit voyages and orders, act on review gates, and retry or skip orders from the browser. Requires Team or Enterprise. - Charts — viewing a chart (
mso chart show) works on every tier; authoring and executing charts (mso chart new, chart-driven routing) requires Pro, Team or Enterprise. - Division packs — free packs apply on every tier; non-free packs and remote signed packs require Team or Enterprise.
- Security tooling — pretool secret detection, posttool OWASP/CWE scanning, branch and path guards that keep each crew inside its own worktree, an MCP server registry with per-server risk levels, and
mso security scan/mso security review. - Audit log — hash-chained JSONL of every privileged action, with optional external anchoring (S3 Object Lock, S3-compatible storage, or Azure Immutable Blob). Switched on per workspace with
mso init --enterprise. - Governance — signed policy bundles. Authoring and signing a bundle (
mso policy keygen|init|lint|sign) works on every tier; publishing, pinning a trusted key and enabling governance on a device require Team or Enterprise;mso policy pulland enforced governed mode (governance.governedMode) are Enterprise only. Guard break-glass overrides are only audited and capability-checked in governed mode, so they are Enterprise-only in practice. - Artefact-repo mode — move Maestro's planning and state artefacts out of your product repo into a dedicated repo. Solo mode works on every tier; shared mode, for several engineers dispatching against one artefact repo, requires Team or Enterprise.
- Slack bridge — mobile control via Slack commands.
- Persona system — swap the vocabulary and tone without changing any code. Five built-in personas ship: Maestro (default), Nautical, Wildlife Reserve, Field Ops, and Imperial Command — or define your own.
Who it's for
Senior individual developers, staff engineers, and small high-velocity teams who already use Claude Code daily and want to scale their throughput without giving up control of their codebase, prompts, or model spend. Teams that want the Hub, shared artefact repos, or signed governance should look at the Team and Enterprise tiers.
If you have not used Claude Code yet, install it first — Maestro assumes Claude Code is already on your machine.