Add command/container execution support to sandbox.mcp configuration

[Copilot]

Enables custom command or container-based execution for the MCP gateway, matching the configuration pattern used for MCP server tools.

Changes

Core types and parsing

• Added Command field to MCPGatewayConfig struct
• Updated parseMCPGatewayTool() to extract command configuration
• Extended JSON schema with command field and updated descriptions

Gateway step generation

• Refactored generateMCPGatewayStartStep() into three execution modes:
  • generateContainerStartCommands() - Docker-based execution with full container control
  • generateCommandStartCommands() - Custom command with args and environment
  • generateDefaultAWMGCommands() - Existing awmg binary fallback

Validation

• Enforces mutual exclusivity between command and container
• Restricts entrypointArgs to container mode only

Testing

• 14 new tests covering all execution modes and validation rules

Usage

Container mode with full Docker configuration:

sandbox:
  mcp:
    container: "ghcr.io/githubnext/gh-aw-mcpg:latest"
    args: ["--rm", "-i", "-v", "/var/run/docker.sock:/var/run/docker.sock"]
    entrypointArgs: ["--routed", "--listen", "0.0.0.0:8000", "--config-stdin"]
    port: 8000
    env:
      DOCKER_API_VERSION: "1.44"

Custom command mode:

sandbox:
  mcp:
    command: "/usr/local/bin/mcp-gateway"
    args: ["--port", "9000", "--verbose"]
    env:
      LOG_LEVEL: "debug"

Default mode (no command/container) continues to use awmg binary as before.

Original prompt

Add support in configuration of sandbox.mcp to configure a command or container that execute. The data structure is similar to configuring a mcp server tool: command, args, container, env, ... When this configuration is selected, the compiler generates a step that invokes the tool configure to launch the gateway.

sandbox:
  mcp:
     container: ...
     command: ... # or conatiner, not both
     args: ...
     entrypointArgs: ... # if container
     env:
       ...: ...

For example:

sandbox:
  mcp:
    container: "ghcr.io/githubnext/gh-aw-mcpg:latest"
    args:
      ["--rm", "-i", "-v", "/var/run/docker.sock:/var/run/docker.sock", "-p", "8000:8000", "--entrypoint", "/app/flowguard-go"]
    entrypointArgs: 
      ["--routed", "--listen", "0.0.0.0:8000", "--config-stdin"]
    port: 8000
    env: 
      "DOCKER_API_VERSION": "1.44" 

The env should be populated from the secrets required by the mcp servers.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.