MCP Code Executor | linuxserver.lan

Request/Response Flow

Dispatch Sequence

Claude Code Invokes MCP Tool

User asks Claude a question. Claude decides to consult an external AI and calls dispatch_to_reviewboard with a prompt and target.

mcp__code-executor__dispatch_to_reviewboard({
  "prompt": "Review /workspace/reviewboard/server.py for security issues",
  "target": "gemini",
  "timeout": 900
})
                        

MCP Server Receives Request

The MCP code-executor server (localhost:9091) receives the request via stdio transport and parses the JSON parameters.

// executor.ts - /reviewboard/dispatch endpoint
const SWARM_NODES = {
  gemini: 'http://reviewboard-gemini:8080',
  codex: 'http://reviewboard-codex:8080',
  claude: 'http://reviewboard-claude:8080'
};
                        

HTTP POST to Swarm Node

MCP server sends HTTP POST to the target container via Docker's internal DNS (mcp-net network).

POST http://reviewboard-gemini:8080/execute
Content-Type: application/json

{
  "prompt": "Review /workspace/reviewboard/server.py...",
  "working_dir": "/workspace"
}
                        

FastAPI Adapter Executes CLI

The FastAPI adapter inside the container spawns the CLI tool (gemini/codex/claude) with the prompt and captures output.

# adapter.py - gemini example
proc = await asyncio.create_subprocess_exec(
    "gemini", "-p", prompt,
    stdout=asyncio.subprocess.PIPE,
    stderr=asyncio.subprocess.PIPE,
    cwd=working_dir
)
stdout, stderr = await proc.communicate()
                        

AI Reads Files & Responds

The AI model reads files from /workspace (if requested), processes the prompt, and generates a response. The adapter captures the output.

JSON Response Returned

The response flows back through the chain: Adapter -> MCP Server -> Claude Code CLI. Claude presents the result to the user.

{
  "target": "gemini",
  "success": true,
  "result": "After reviewing server.py, I found the following...",
  "metrics": {
    "executionTime": 12500,
    "model": "gemini-2.5-flash"
  }
}
                        

MCP API Endpoints

POST /reviewboard/dispatch

Dispatch a prompt to an AI Board node for execution.

// Request Body
{
  "prompt": string,     // The task/question to send
  "target": "gemini" | "codex" | "claude",
  "timeout": number,    // Seconds (default: 900)
  "working_dir": string // Default: /workspace
}

// Response
{
  "success": boolean,
  "result": string,
  "error": string | null,
  "metrics": { "executionTime", "model" }
}
                

GET /reviewboard/health

Check health status of all swarm nodes.

// Response
{
  "swarm": {
    "gemini": { "status": "healthy", "node_type": "gemini" },
    "codex": { "status": "healthy", "node_type": "codex" },
    "claude": { "status": "healthy", "node_type": "claude" }
  },
  "healthyNodes": 3,
  "totalNodes": 3
}
                

Usage Patterns

Code Review

// Ask Gemini to review code
dispatch_to_reviewboard({
  prompt: "Review the code in /workspace/myproject/src/ for security issues. Read the files first.",
  target: "gemini"
})
                    

Second Opinion

// Get alternate perspective
dispatch_to_reviewboard({
  prompt: "Gemini suggested using Redis. What's your perspective on Redis vs PostgreSQL for session storage?",
  target: "codex"
})
                    

Internet Research

// Gemini has web access
dispatch_to_reviewboard({
  prompt: "Search for the latest Docker security best practices in 2025",
  target: "gemini"
})
                    

Deep Analysis

// Complex reasoning with Claude
dispatch_to_reviewboard({
  prompt: "Analyze the architecture in /workspace/reviewboard and suggest improvements for scalability",
  target: "claude"
})
                    