Skip to main content
Glama
seedpod

seedpod

Official
by chlorofil-os

Seedpod

Your LXC machines, and the Docker running inside them, in one tree.

Seedpod is a VS Code sidebar and an MCP server for people who run their dev environments as LXD system containers: a machine per branch, per client, per experiment, each one with its own Docker Compose stack inside.

Why Seedpod

LXD system containers are a great way to run many full dev machines on one host. They boot in seconds, clone in seconds on btrfs or ZFS, and run systemd and Docker like a VM, at a fraction of the cost. But the tooling stops at the machine boundary:

  • Container Tools only sees the host's Docker. The compose stacks inside your LXD machines are invisible to it, so checking on them means lxc exec <machine> -- docker ps, again and again, one machine at a time.

  • lxc and docker are two separate worlds. Is web down because the container stopped, or because the machine around it did? You end up asking both, by hand.

  • Agents get a shell or nothing. Giving an AI assistant raw lxc access means it can delete anything. Giving it none means it can't help.

Seedpod puts both layers in one place:

  • One tree. Machine → compose project → container, with live status, IPs and uptime, refreshed as LXD reports changes.

  • One right-click. Start, stop, open a shell, read logs or snapshot at whichever level is broken, with a confirmation that names the target before anything destructive.

  • The same model for AI clients. The MCP server gives assistants structured tools instead of a shell. Machines you list as protected can't be changed, there's a read-only mode, and there are no delete or restore tools at all.

Related MCP server: VPS-Guardian-MCP

The extension

  • Containers: LXD instances grouped by project, with status, OS, IP and uptime. Expanding a running machine lists the Docker containers inside it, grouped by compose project, plus its snapshots.

  • Images, Networks and Storage views (collapsed by default).

  • Right-click actions:

    • Machines: start / stop / restart / freeze, open a shell, show the log, inspect, snapshot / restore / delete, copy IP, open a Remote-SSH window, delete.

    • Docker containers: start / stop / restart / remove, view logs, attach a shell, inspect.

    • Compose groups: start / restart / stop all.

  • Refreshes live from lxc monitor. Docker inside expanded machines is polled.

Every destructive action asks first, in a modal that names the target.

Settings

Setting

Default

seedpod.cliPath

lxc

Path to the lxc binary. Setting it to incus is best effort and untested: snapshot verbs are mapped, other output is assumed compatible with LXD 5.21.

seedpod.docker.pollInterval

10

Seconds between Docker refreshes for expanded machines. 0 turns polling off.

seedpod.networks.showUnmanaged

false

Also list host bridges LXD does not manage (docker0, br-*).

seedpod.remoteSsh.user

ubuntu

User "Open in Remote-SSH Window" connects as.

seedpod.remoteSsh.folder

""

Folder to open, e.g. ~/workspace; ~ resolves inside the machine. Empty opens an empty window.

seedpod.remoteSsh.machineSettings

{}

VS Code settings written into the machine's ~/.vscode-server/data/Machine/settings.json before connecting. Other keys there are kept; a file with comments is left alone.

For example, to open ~/workspace and see git worktrees that live one folder deeper than the repos:

"seedpod.remoteSsh.folder": "~/workspace",
"seedpod.remoteSsh.machineSettings": {
  "git.repositoryScanMaxDepth": 2,
  "scm.alwaysShowRepositories": true
}

Notes

  • Only the local daemon is supported. Remote LXD servers are not.

  • Docker is reached through lxc exec <machine> -- docker …, so the host needs no Docker socket access.

  • "Open in Remote-SSH" connects as ubuntu@<ip> (see seedpod.remoteSsh.user). Keys come from your SSH config.

Build

npm install
npm test            # parser and tree tests (node --test)
npm run build       # esbuild → dist/extension.js
npm run package     # → seedpod-<version>.vsix
code --install-extension seedpod-<version>.vsix

MCP server

The same code also ships as a stdio MCP server (dist/mcp.js, built by npm run build). It lets an AI client see and manage the machines and the Docker containers inside them.

Tool

Writes

list_machines

every instance: status, OS, IPv4, uptime, snapshots, protected flag

machine_info, machine_log

expanded config + lxc info, and the LXC log

list_docker

Docker containers inside a machine, grouped by compose project

docker_logs, docker_inspect

bounded logs (no follow), and docker inspect

list_resources

images, managed networks, storage pools

machine_action

✔

start / stop (100 s clean timeout, never forced) / restart / freeze / unfreeze

create_snapshot

✔

new snapshot (validated name)

docker_action, compose_action

✔

start / stop / restart a container or a whole compose project

exec

✔

run an argv inside a machine as root and return the exit code and output

There are deliberately no delete, restore or remove tools.

Env var

SEEDPOD_CLI

path to lxc (default lxc)

SEEDPOD_READONLY=1

register the read-only tools only

SEEDPOD_PROTECT=a,b

machines no write tool (including exec) may touch

To register it with Claude Code:

claude mcp add --scope user seedpod -e SEEDPOD_PROTECT=my-template -- node /path/to/seedpod/dist/mcp.js

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Provides Docker management tools for AI agents, including inspecting containers, images, Dockerfiles, and Compose configurations.
      62 npm
      1
      MIT
    • A
      license
      B
      quality
      B
      maintenance
      Nido is a local QEMU VM manager. Install it on your computer from https://github.com/Josepavese/nido and run nido mcp via stdio, or use npx -y nido-mcp@latest mcp with QEMU installed locally. It manages VMs on that same computer; a server hosted by Glama cannot access your local VM inventory.
      6
      2
      MIT