agentic-coding-container

A Docker image that runs an SSH server with AI coding agents pre-installed, intended for use as a remote development environment. Compatible with the ssh feature of the Claude desktop app allowing for development on a remote machine, for example a kubernetes node with a GPU.

Ships with Claude Code and OpenAI Codex out of the box — either can be toggled off at build time.

What it does

• Runs an OpenSSH server accessible via public key authentication
• Installs Claude Code and Codex CLIs globally via npm (configurable)
• Runs as the ubuntu user
• SSH host keys are persisted via a mounted volume so they survive container restarts

Build

docker build -t agentic-coding-container .

Build arguments

Argument	Default	Description
BASE_IMAGE	ubuntu:24.04	Base image to build from
NODE_VERSION	22	Node.js major version to install
EXTRA_APT_PACKAGES	—	Space-separated extra apt packages to install
EXTRA_NPM_PACKAGES	—	Space-separated extra global npm packages to install
ENABLE_SUDO	false	Grant the ubuntu user passwordless sudo (grants full root access — use with caution)
INSTALL_CLAUDE	true	Install Claude Code CLI
INSTALL_CODEX	true	Install OpenAI Codex CLI

# Custom base image and Node version
docker build --build-arg BASE_IMAGE=ubuntu:22.04 --build-arg NODE_VERSION=20 -t agentic-coding-container .

# Claude Code only
docker build --build-arg INSTALL_CODEX=false -t agentic-coding-container .

# Codex only
docker build --build-arg INSTALL_CLAUDE=false -t agentic-coding-container .

Run

Two mounts are required:

• Host keys volume at /etc/ssh/host-keys — persists SSH host keys across restarts
• Authorized keys file at /etc/ssh/authorized_keys/authorized_keys — public keys allowed to connect

docker run -d \
  -p 22:22 \
  -e ANTHROPIC_API_KEY=sk-ant-... \
  -e OPENAI_API_KEY=sk-... \
  -v host-keys:/etc/ssh/host-keys \
  -v /path/to/authorized_keys:/etc/ssh/authorized_keys/authorized_keys:ro \
  agentic-coding-container

Environment variables

Variable	Default	Description
PORT	22	Port sshd listens on
ANTHROPIC_API_KEY	—	Injected into the ubuntu user's environment for Claude Code (written to .bashrc/.profile in plaintext)
ANTHROPIC_BASE_URL	—	Optional custom Anthropic API base URL (passed through via SSH AcceptEnv)
OPENAI_API_KEY	—	Injected into the ubuntu user's environment for Codex CLI (written to .bashrc/.profile in plaintext)

Connecting

ssh ubuntu@<host>

Desktop app sandboxing

The Claude Desktop app runs in a sandbox that may block outbound SSH connections. If the app cannot reach your server directly, use a local port forward so it connects via localhost instead.

SSH port forward:

ssh -N -L 2222:<host>:22 ubuntu@<host>

Then point the desktop app at localhost:2222.

Persistent forward with autossh:

autossh -M 0 -N -L 2222:<host>:22 ubuntu@<host>

autossh monitors the tunnel and restarts it if it drops. Install via brew install autossh on macOS.

Kubernetes deployment

Volumes

Mount	PVC	Purpose
/home/ubuntu/workspace	workspace-pvc (10Gi)	Project code
/etc/ssh/host-keys	ssh-host-keys-pvc (1Mi)	Stable SSH host keys

Setup

1. Build and push the image

The manifest defaults to ghcr.io/n1mmy/claude-server:main. To use a custom image:

docker build -t your-registry/agentic-coding-container:latest .
docker push your-registry/agentic-coding-container:latest

Then update image: in k8s-manifest.yaml to match.

2. Set your SSH public key

kubectl create secret generic ssh-authorized-keys \
  --from-file=authorized_keys=$HOME/.ssh/id_ed25519.pub \
  -n claude-workspace

3. Set your API keys

kubectl create secret generic ai-credentials \
  --from-literal=anthropic-api-key=sk-ant-... \
  --from-literal=openai-api-key=sk-... \
  -n claude-workspace

4. Deploy

kubectl apply -f k8s-manifest.yaml

5. Get the SSH address

kubectl get svc agentic-coding-ssh -n claude-workspace
# Note the EXTERNAL-IP

For local clusters without LoadBalancer:

kubectl port-forward svc/agentic-coding-ssh 2222:22 -n claude-workspace
# Then SSH to localhost:2222

6. Connect

Host agentic-coding
    HostName <EXTERNAL-IP>
    User ubuntu
    Port 22
    IdentityFile ~/.ssh/id_ed25519

GPU nodes

To request a GPU, add to the container resources in k8s-manifest.yaml:

resources:
  limits:
    nvidia.com/gpu: "1"

And add a node selector or toleration for your GPU node pool.