claude-spawn

Spawn containerized Claude Code agents — locally, on Kubernetes, or on remote servers. The orchestrator plugin.

Install

From the NodeNestor marketplace:

/plugin marketplace add https://github.com/NodeNestor/nestor-plugins
/plugin install claude-spawn

Or manually:

git clone https://github.com/NodeNestor/claude-spawn.git ~/.claude/plugins/claude-spawn

Usage

Just ask Claude naturally:

"Spawn an agent to fix the failing tests in this repo"

"Create an agent on my k8s cluster with GPU to train the model"

"Spin up an agent on the staging server to run the deploy"

"How are my agents doing?"

Claude handles the tool calls behind the scenes.

Targets

Target	Where it runs	What you need
local	Local Docker (default)	Docker
k8s	Kubernetes cluster	kubectl
remote	Remote Docker host	SSH access

"Spawn an agent locally to fix tests"
"Spawn an agent on k8s node gpu-node-1 with GPU"
"Spawn an agent on ssh://deploy@prod-server"

What it does

• Multi-target — local Docker, Kubernetes pods, or remote Docker hosts
• Optional desktop — Xvfb + Chromium + noVNC for browser tasks
• Optional GPU — NVIDIA runtime / K8s GPU resources
• Credential forwarding — auto-inherits host auth (local: mount, k8s: secret, remote: mount)
• Plugin support — auto-install any NodeNestor plugin stack
• SSH + tmux — connect to any agent directly

Plugins (NodeNestor stack)

Plugin	Description
rolling-context	Context compression proxy for long sessions
knowledge-graph	Persistent cross-session memory
workflows	Agent controller — YAML workflows, cron, GitHub polling
autoresearch	Auto-research any codebase on first session
worktrees	Parallel experimentation via git worktrees

MCP servers (third-party)

Server	Description	Default
playwright	Browser automation and testing	yes
context7	Live docs for any library	yes
github	GitHub API (needs GITHUB_TOKEN)	no
fetch	HTTP requests	no
postgres	PostgreSQL (needs POSTGRES_CONNECTION_STRING)	no
sqlite	SQLite databases	no

Tools

Tool	Description
spawn_agent	Create a new agent (any target)
list_available	See all plugins, MCP servers, and targets
list_agents	List agents across all targets
agent_status	Get agent status and ports
agent_logs	Read agent output
agent_exec	Run a command in an agent
stop_agent	Stop and remove an agent

Auth

By default, agents inherit credentials from the host. To use custom auth:

• api_key — custom Anthropic API key
• api_url — custom API base URL (e.g. a proxy)

When rolling-context is in plugins, the proxy chain auto-configures:

Claude Code -> rolling-context (:5588) -> api_url (or api.anthropic.com)

Requirements

• Docker (for local/remote targets)
• kubectl (for k8s target)
• Python 3.10+ (for the MCP server)