Skip to content

Turn a list of servers
into one cluster.

mm is a Docker Swarm replacement that runs your Compose project across every host you list. The nodes get an encrypted private network, service names that resolve on all of them, rolling deploys that roll back, shared volumes, and one merged log. Coding agents operate it within limits you set.

.config/mm.yaml
targets:
production:
nodes:
- {name: app-1, host: 203.0.113.11}
- {name: app-2, host: 203.0.113.12}
- {name: db-1, host: 203.0.113.13}

A complete cluster target. The mesh, the network ranges, and the storage node have defaults.

mm -e production · ~/code/acmetour
mmacmeproductionremote · production·docker compose
5/5 running no issueslive
Workspace
Health
Security
Automation
Setup
Services
5/5
running
Routes
0
of 5 services routed
Issues
0
all checks passed
HTTPS
trusted
2 hosts
Workspace2 environments
Root~/code/acme
Runs onremote · production
Configvalid
Runtimedocker compose
Engine28.1.1
Env deliveryenv-file · development-default
Environments3 · 2 remote
Needs attention
Everything looks good

This terminal is interactive. Click it, then pressuenterjl2?

.config/mm.yaml
targets:
production:
nodes:
- {name: app-1, host: 203.0.113.11}
- {name: app-2, host: 203.0.113.12}
- {name: db-1, host: 203.0.113.13}

A complete cluster target. The mesh, the network ranges, and the storage node have defaults.

Decide what a coding agent may do
in each environment.

mm agent setup connects Claude Code, OpenCode, Codex, or Cursor to mm's MCP server. An agent can run jobs and commands in local, deploy to staging, and only read nodes, status, and logs on the production cluster.

  • An agent reads the cluster the way you do

    Nodes, the mesh, name resolution, deployments, and logs come back as the same structured facts the CLI prints, so an agent works from what the hosts report.

  • One reviewed block sets the limits

    The agents block in .config/mm.yaml gives each environment a level and, separately, exec. It takes effect only after you trust it on your machine, and again after every edit.

  • Past the limit, the agent hands you the command

    A call outside the grant changes nothing. It returns the exact mm command, and the agent asks you to run it in your own terminal.

  • Every node enforces the limits itself

    An agent connects with its own key and gets only what the deployed commit allows. mm refuses to run inside a coding tool's shell, and your own key stays behind a passphrase, so an agent cannot pick it up with ssh.

.config/mm.yamltrusted
agentschange a level
  • local
  • staging
  • production
agents:
"*": view
local: { level: deploy, exec: true }
staging: deploy
production: view
agent session
  • agentmm_read nodes production
    mm3 nodes ready · all on 3f2c9a1e5b7d
  • agentmm_job_apply run seed local
    mmstarted · seed attempt 1/1
  • agentmm_deploy_apply plan 7c1e9a4f staging
    mmapplied · api 3/3 healthy
  • agentmm_exec api -- rails db:migrate production
    mmhandoff · needs exec, you run:mm --environment production service exec api -- rails db:migrate

Everything between the nodes is set up for you.

mm target setup puts the container engine, the mmd daemon, and a WireGuard mesh on every host. After that a cluster has four things to think about: nodes, services, networks, and volumes.

The nodes talk to each other over WireGuard.

Setup gives every node an address on a mesh that mmd owns, and you configure nothing for it. One UDP port has to be open between the nodes, a node's private key never leaves it, and traffic between containers on different nodes travels inside the mesh.

The mesh between the nodes
mesh
mm -e production target mesh status
Target production: mesh between 3 node(s)
app-1 mm0 up 10.90.0.1 UDP 51820 mtu 1420 key 4kT9wq0aYx1B…
PEER ENDPOINT HANDSHAKE RECEIVED SENT
app-2 203.0.113.12:51820 34 s ago 18 MiB 21 MiB
db-1 203.0.113.13:51820 12 s ago 62 KiB 66 KiB
app-2 mm0 up 10.90.0.2 UDP 51820 mtu 1420 key Zr81nQe0pL2c…
PEER ENDPOINT HANDSHAKE RECEIVED SENT
app-1 203.0.113.11:51820 34 s ago 21 MiB 18 MiB
db-1 203.0.113.13:51820 9 s ago 31 KiB 31 KiB
db-1 mm0 up 10.90.0.3 UDP 51820 mtu 1420 key b0Hq7TtW3xNe…
PEER ENDPOINT HANDSHAKE RECEIVED SENT
app-1 203.0.113.11:51820 12 s ago 66 KiB 62 KiB
app-2 203.0.113.12:51820 9 s ago 31 KiB 31 KiB
Every node has a recent handshake with every other.

There is no scheduler and no consensus store to run. A node that goes down comes back with its own services, and nothing moves to another node in the meantime. When a node goes down

Develop on your laptop with the config you deploy.

Keep container commands, health checks, and volumes in Compose. mm adds trusted local HTTPS, secrets, source mounts, and jobs around them, and shows you each change before it happens.

mm deploy plan

Nothing changes until you have read the plan.

Builds, deployments, and shared services each produce a plan and print its hash. Apply takes the hash of the plan you reviewed. If the config, a secret, or the target moved in between, mm stops and asks you to plan again.

How plans work
$mm deploy plan
~apiimage5d0c… → 9b1e…
~webenvAPI_URL
=workerunchanged
plan7c1e9a4f
$mm deploy apply --plan 7c1e9a4f --yes
applied
mm dev tls

Real HTTPS on your laptop.

Wildcard .test DNS and a local certificate authority your machine trusts. Every routed service gets a hostname. No hosts-file entries, no browser warnings.

https://web.acme.mm.test
  • ↗api.acme.mm.test8080
  • ↗web.acme.mm.test3000
  • Trusted by your user certificate store
Local HTTPS
mm secret

Secrets never touch the repo.

Config holds references. Values live in a separate store: local, OpenBao, or Vault. Plans include the reference and never the value.

DATABASE_URLconfigpostgres://postgres:5432/acme
SESSION_SECRETsecret••••••
LOG_LEVELlocal overridedebug
Environments and secrets
.config/mm.yaml

Your Compose files stay exactly as they are.

Image repositories, environments, secret references, and routes go in one file next to them. Commands, health checks, and volumes stay in Compose.

modules:
  backend:
    path: backend
    build:
      image: ghcr.io/acme/api
    services:
      api: { stack: app, role: http }
What goes where
mm team

A file in the repo decides who reaches each host.

The users file gives each person and each CI key a level per environment. A change takes effect when it is merged and deployed, and every node applies it.

alicedeploy
bobview
ciautomationdeploy
Remote access
mm job

Jobs and schedules

Declare migrations, cleanups, and recurring work next to your services. Run, retry, and cancel them from the CLI or the terminal interface.

mm dev dashboard

A dashboard and a mailbox

A local web dashboard shows status across projects, and Mailpit catches every email your services send in development.

mm dev mounts

Live code in running containers

Development mounts map module source into running containers, with presets that keep dependencies and protected files out of the way.

Install mm with one command.

The installer fetches a verified release and puts mm on your machine. Workstation setup is a separate step that shows its plan and asks before it changes anything.

install
curl --proto '=https' --tlsv1.2 -LsSf https://get.modmon.org/install.sh | sh
mm dev setup# local trust, ingress, dashboard, Mailpit
mm# open the guided interface
--version pin a release--install-dir choose where--with-daemon also install mmd
you need

x86-64 or ARM64

  • curl or wget, and tar
  • Docker with Compose, or Podman Compose
  • systemd, for local HTTPS and the dashboard
the installer
  • Downloads the release for your platform over HTTPS.
  • Checks the archive's size and SHA-256 against the release manifest before extracting it.
  • Installs mm in ~/.local/bin. No sudo, and no toolchain to build.
  • Replaces an existing mm atomically, and refuses to downgrade unless you pass --force.