Skip to main content
Glama
adamdude828

mcp-server-docker

by adamdude828
README.md
# MCP Server: Docker Command Runner

This MCP (Model Context Protocol) server provides a secure interface for running commands inside Docker containers. It acts as a privileged sidecar that can execute arbitrary commands within specified Docker Compose service containers.

## Features

- STDIO-based MCP transport for integration with Claude and other MCP clients
- Execute commands in any Docker Compose service container
- Real-time capture of stdout/stderr output
- Secure container allowlist configuration
- Configurable timeouts for long-running commands
- Clear error messages for common Docker issues
- Minimal dependencies and secure by design

## Installation

### Via NPX (Recommended)

Run the MCP server directly without cloning:

```bash
npx mcp-server-docker
```

Or install globally:

```bash
npm install -g mcp-server-docker
mcp-server-docker
```

### As a Docker Service

Add the following to your `docker-compose.yml`:

```yaml
services:
  mcp-docker:
    build: ./mcp-server-docker
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    ports:
      - "3001:3000"  # Expose MCP server
    environment:
      - COMPOSE_PROJECT_NAME=${COMPOSE_PROJECT_NAME}
      - DEFAULT_SERVICE=app
      - COMPOSE_FILE=docker-compose.yml
      - PORT=3000
    networks:
      - your-network
```

### For Local Development

```bash
cd mcp-server-docker
npm install
npm run build
```

## Configuration

The server accepts the following environment variables:

- `ALLOWED_CONTAINERS`: Comma-separated list of allowed service:container pairs (e.g., "app:myapp_container,db:mydb_container")
- `DEFAULT_SERVICE`: Default service to run commands in (default: "laravel_app")
- `COMMAND_TIMEOUT`: Command timeout in milliseconds (default: 300000)

## MCP Tool: run_command

The server exposes a single tool called `run_command`:

### Input Schema

```json
{
  "command": "string (required) - The command to execute",
  "service": "string (optional) - Docker service name"
}
```

### Example Usage

```json
{
  "command": "npm test",
  "service": "frontend"
}
```

### Response Format

The tool returns the command output with the following structure:
- Standard output (if any)
- Standard error (if any, prefixed with [stderr])
- Exit code

## Usage

The server uses STDIO transport for MCP communication. When run with `npx mcp-server-docker`, it will:

1. Parse environment variables for allowed containers
2. Start the MCP server listening on stdin/stdout
3. Log startup information to stderr
4. Wait for MCP protocol messages

## Security Notes

- This server requires access to the Docker socket - ensure Docker is running and accessible
- No command filtering is applied - relies on container isolation for security
- Commands timeout after 5 minutes by default
- Only allowed containers (configured via ALLOWED_CONTAINERS) can be accessed

## Development

```bash
# Run in development mode
npm run dev

# Build TypeScript
npm run build

# Run tests
npm test
```

## Troubleshooting

### Common Errors

1. **"Cannot connect to Docker daemon"**: Ensure Docker is running and the socket is mounted
2. **"Service not found"**: Check that the service name exists in your docker-compose.yml
3. **"Command timed out"**: Command exceeded 5-minute timeout, consider breaking it into smaller operations

TDQS

B3/5.0

Scored across 1 tool

Disambiguation5/5

Only one tool exists, so there is no possibility of confusion between tools. The tool's purpose is clear and distinct from any non-existent others.

Naming Consistency5/5

The single tool name 'run_command' follows a clear verb_noun pattern. With only one tool, there are no inconsistencies to evaluate.

Tool Count2/5

The server is named for Docker but has only one tool, which is clearly too few for Docker-related operations. A single tool feels minimal and restricts the server's utility.

Completeness1/5

The tool surface is severely incomplete for a Docker server. It only supports running commands inside containers, with no tools for managing containers, images, networks, or volumes. This will cause agent failures in most Docker workflows.

Maintenance

ActivityInactive
ResponsivenessUnresponsive