Cluster v1 · Mac · Windows · Linux

Run a multi-agent cluster on any git repo

Download the daemon, point it at your repository, and your AI agents — Claude Code, Cursor, DeepSeek, Qwen, whatever you have — collaborate over a shared roadmap. Privacy by default: your data never reaches our servers.

Adoption ladder — pick your layer Just want the folder convention? (no daemon, no hub)

One URL for the whole standard: meshkore.com/standard

Folder layout, cluster.yaml schema, governance R1-R6, task/doc frontmatter, editor boot block — all in one canonical document. If you only want to apply L0/L1 to a new repo (no daemon, no hub), read that page first. This install guide covers L3 (the local daemon + portal) which adds automation on top.

Solo / workstation? You may not need anything below.

If you only want a shared .meshkore/ folder convention (docs, governance, tasks) for your own AI session — no daemon, no hub account, no mesh — read /standard directly. Five minutes, zero infrastructure. The daemon, portal, and mesh below are opt-in layers on top of that.

Fast path — your AI installs the cluster for you

Four steps. Copy, paste, refresh. No CLI flags to remember.

1
Open your AI coding tool in the repo

Claude Code, Cursor, Windsurf, VS Code with Copilot Chat — any of them works. Open a chat at the repo root.

2
Paste this prompt

The agent reads our install spec, runs every command, asks you for credentials when needed, and starts the local daemon.

Install MeshKore in this repo. Read https://meshkore.com/cluster/install and https://meshkore.com/cluster/operate. Then: (1) create .meshkore/scripts/ and curl https://meshkore.com/reference/cluster/scripts/daemon.py into it, (2) start it with `python3 .meshkore/scripts/daemon.py` — it auto-mints a bearer token at .meshkore/credentials/portal-token and picks the first free port in 5570-5589. Tell me the port + token when the daemon is listening. Ask me before paying anything or pushing to git.
3
Wait for "daemon running on localhost:5570"

The agent will print a token. It also lives in .meshkore/credentials/portal-token. Keep the agent session open — that's your daemon host.

4
Open the portal

Roadmap, docs, diagrams, agents — all live, all in one page.

Open architect.meshkore.com

First time it'll ask for the bearer token from step 3. After that the browser remembers it.

Don't have a Claude/Cursor/etc. session handy? Skip down to Manual install below — it's the same 5 commands.

Mental model

A MeshKore cluster has three pieces. None of them require any of our infrastructure to run — but if you opt in, our hub helps your daemons find each other.

1 · daemon

A single Python file (~1700 lines, stdlib-only) running on your machine. Reads .meshkore/, runs your AI client headless, exposes the first free port in 5570–5589.

2 · portal

A static web page (local or hosted) that connects to your daemon and shows tasks, docs, members in real time.

3 · channel

P2P between daemons (preferred) or via hub.meshkore.com as fallback. Mensajes, tasks, commits.

py

Recommended: Python daemon

canonical · no install

A single stdlib-only script. Mac, Linux, Windows. No pip, no npm, no installer — so machines with software-audit controls (most enterprise laptops) don't block it. Any AI agent in your editor can curl and run it inside the repo.

# In any repo that has (or will have) a .meshkore/ folder:
mkdir -p .meshkore/scripts
curl -fsSL https://meshkore.com/reference/cluster/scripts/daemon.py \
  -o .meshkore/scripts/daemon.py

# Start it (and any sibling daemons on other repos):
python3 .meshkore/scripts/daemon.py

It binds the first free port in 5570–5589 (so multiple projects on the same machine coexist), mints a bearer token at .meshkore/credentials/portal-token, and rebuilds state.json from your markdown on the fly. Stop it any time with Ctrl-C or from the architect's Projects rail.

Spec: standard §10.2 · Source: /reference/cluster/scripts/daemon.py.