Maestro coordinates up to five parallel Claude Code squads against a shared workspace. Specialised roles, dependency-ordered work, quality gates, security scanning. Bring your own key.
Maestro is a Python CLI. It runs on your machine, talks to your Anthropic key, and orchestrates Claude Code sessions you already have access to.
One pip install maestro-fleet from PyPI, with no GitHub credentials needed. Activate your licence, then scaffold a workspace with mso init.
A voyage is a sprint. Drop orders into .mso/queues/orders/. Each order names a role: PLN, BLD, TST, DOC, SEC, REL.
Up to five squads run in parallel. Quality gates between roles. Security scans on every write. Real-time status in your terminal.
A real quickstart session: install, check your licence tier, scaffold, create a voyage and an order, and dispatch your first crews.
Maestro ships five built-in personas. Each persona changes how squads address the operator, the terminology in prompts, and the flavour of CLI output. The orchestration mechanics are identical underneath.
The classic Maestro experience. You are the Captain; Claude is your Quartermaster. Squads are crews; deployments are voyages.
| Operator | Captain |
| Lead agent | Quartermaster |
| Squads | Crews |
| Sprint | Voyage |
Set your persona: run mso settings set persona nautical, or export MAESTRO_PERSONA=nautical to override it for a single process. mso setup also offers a persona step. The installation and configuration guide has the full reference.
Six bound roles (Planner, Builder, Tester, Documenter, Security, Releaser), each running on the right model tier (Haiku for read-only security review; Opus for planning and release decisions). Squads dispatch in waves: planning waits on nothing, building waits on planning, QA waits on building. Five Claude Code sessions running concurrently against the same repo without stepping on each other.
$ mso dispatch --max-crews 5Structured acceptance criteria attach to every order. The Tester validates each one: file-exists, manual judgement, or an arbitrary shell command exit code. The Releaser will not recommend a merge unless every gate has passed, and the merge itself stays yours to make. Pretool secret detection blocks AWS keys, GitHub PATs, OpenAI tokens before they hit a commit. Posttool OWASP/CWE scanning runs against every Security review.
$ mso security scan --voyage VOY-0026Maestro is a Python CLI installed via pip. There is no Maestro server. Your code never crosses the network to us. Your Anthropic API key is yours; you pay Anthropic directly for token usage. The audit trail, voyage state, and order history all live in .mso/ next to your code. Air-gapped operation supported.
$ mso data-flow --mode air-gappedInstall and dispatch get you started. These are what running a fleet actually looks like.
mso doctor is a read-only self-check across the dispatcher, crews, orders, voyages and machine hygiene. Every finding proposes a command; nothing is mutated for you. It works with the dispatcher stopped, which is exactly when you need it.
mso pause drains: no new work is claimed, running crews finish their turn, then the dispatcher exits cleanly. Nothing is killed, salvaged or retried.mso resume picks straight back up.
GitHub or Bitbucket for source, GitHub Actions or Jenkins for CI, GitHub Issues or Jira for tracking. A provider you have not configured is reported as a named skip, never a silent pass.
mso lead start runs a per-workspace background agent that survives your terminal closing. It patrols on a cadence and handles two reversible chores on its own, logging each one. It never merges a PR or touches production code unattended.
mso hub start serves the dashboard from the pip package itself: no Docker and no Node. The Hub and the resident agent are licensed features; talk to us about an edition.
Other coding agents in your toolbox. Where each fits, and where Maestro fits instead.
Single-developer, in-IDE pairing. Brilliant for inner-loop edits.
Maestro runs five orchestrated squads against shared workspace state: not in-IDE, complementary.
Cloud-hosted autonomous agents. Their servers, their orchestration, your code goes to them.
Maestro runs entirely on your machine with your key. No SaaS in the loop.
Single session, no orchestration, no role separation, no quality gates.
Maestro is a productivity layer on top of Claude Code: five sessions, six roles, gates between them.
Full side-by-side comparisons
A bug was found mid-session. The LEAD hand-applied a fix in five minutes. A Maestro crew independently fixed the same bug, more robustly and with tests. The crew won on every quality dimension.
Most coding agents trade control for convenience. Maestro is the opposite trade.
You sign up with Anthropic directly. We never see your tokens, and we take no margin on your model usage.
Pip-installed CLI. No Maestro server in the request path. Your repo, your shell, your machine, from start to finish.
Restrictive and air-gapped modes refuse outbound HTTP outside your allowlist. mso data-flow renders every host the framework would reach for.
You always pay Anthropic directly for the model usage. Maestro's licence covers the orchestration layer.
Free for solo developers. 1 project, 2 squads, 4 roles (INV / PLN / BLD / TST), pretool secret detection, voyage and order schema, chart visualization.
Get started free →Unlimited projects, all 5 squads and 6 roles, a resident LEAD sub-agent that patrols your fleet and handles routine housekeeping on its own, security tooling, MCP registry, Slack bridge, audit anchoring, typed chart routing with bounded revision cycles, shared team workspace, SSO / SLA on Enterprise.
Talk to us →Published on PyPI, no GitHub credentials needed. You'll be dispatching squads inside 90 seconds.