multibase 🏗️

Run multiple Supabase projects on one host — share the heavy services, keep projects isolated.

Running several Supabase projects locally? Each supabase start spins up a full stack — Postgres, Kong, Studio, Realtime, Logflare, Storage, Auth, Rest, Inbucket, Edge Runtime, Vector... 10+ containers per project.

With 3 projects that's 30+ containers and ~4.4 GB RAM.

multibase shares the infrastructure services and runs only the project-specific parts per project. Same isolated databases, same Supabase CLI compatibility — but half the RAM.

How it works

Without multibase:
  Project A ── 10 containers ──┐
  Project B ── 10 containers ──┤  30 containers
  Project C ── 10 containers ──┘  ~4.4 GB RAM

With multibase:
  ┌─ Shared ──────────────────┐
  │  Postgres (3 databases)    │
  │  Kong (API gateway)        │
  │  Studio (1 instance)       │
  │  Realtime (1 instance)     │
  │  Logflare (1 instance)     │
  │  Storage (1 instance)      │
  └────────────────────────────┘
  ┌─ Per project ─────────────┐
  │  GoTrue (Auth)            │
  │  PostgREST (REST API)     │
  │  Edge Runtime (Functions) │
  │  Inbucket (Email)         │
  └────────────────────────────┘
  ≈ 15 containers, ~2.0 GB RAM

Installation

pip install multibase

Or from source:

git clone https://github.com/moinsen-dev/multibase.git
cd multibase
pip install -e .

Quick start

# 1. Init a project
multibase init my-project --port-base 54320

# 2. Edit multibase.yaml to add more projects
#    (or use `multibase add`)

# 3. Start everything
multibase up

# 4. Check running services
multibase ps

# 5. See port mapping
multibase ports

# 6. Stop
multibase down

Add a project

multibase add watch-now --port-base 55320
multibase up  # starts only the new project's services

Port allocation

Each project gets a port base — all services are base + offset:

Service	Offset	Example (base 54320)
Kong	+1	54321
Auth	+2	54322
REST	+3	54323
Studio	+4	54324
Inbucket	+5	54325
Edge Runtime	+6	54326

Remote access (with Tailscale)

multibase auto-detects your Tailscale IP and prints remote URLs:

$ multibase up
📡 Access URLs:
   tinideas:
     API:    http://localhost:54331
     Studio: http://localhost:3000?project=tinideas | http://100.64.0.1:3000?project=tinideas

AI Control (MCP Server)

multibase comes with a built-in MCP server that lets AI agents manage your Supabase cluster.

stdio mode (local AI agents)

Pipe multibase's MCP server to Claude Code, Hermes, or any MCP-compatible agent:

# In Claude Code / Cursor / etc:
multibase mcp --transport stdio

# The agent can now:
# - Add projects   → multibase_add_project("watch-now", 55320)
# - Start/stop     → multibase_up() / multibase_down()
# - Check status   → multibase_status()
# - View config    → multibase_list_projects()

HTTP mode (remote AI agents)

Run the MCP server over HTTP for remote control via Tailscale:

multibase mcp --transport http --port 8085

Then configure your AI agent with:

{
  "mcpServers": {
    "multibase": {
      "url": "http://100.64.0.1:8085"
    }
  }
}

Available MCP tools

Tool	Description
multibase_init	Create new multibase config
multibase_add_project	Add Supabase project
multibase_remove_project	Remove project
multibase_up	Start all services
multibase_down	Stop all services
multibase_status	Running containers + RAM
multibase_list_projects	Configured projects
multibase_show_config	Full compose config
multibase_show_kong_config	Kong declarative config

Requirements

• Docker / OrbStack
• Python 3.10+
• Optional: Tailscale for remote access

License

MIT — see LICENSE.

Contributing

PRs welcome! The project is in early alpha — lots of room for improvement:

• Kong declarative config generation with per-project routes
• multibase db create to initialize per-project databases
• Auto-migration of supabase CLI projects
• Per-project JWT secret generation
• CLI completions
• Tests