Claude Desktop Commander MCP

Instructions

Implementation Reference

• src/tools/execute.ts:5-28 (handler)

  The core handler function for the 'execute_command' tool. Parses input arguments, validates the command, executes via terminalManager, and formats the response with PID and initial output.

  export async function executeCommand(args: unknown) {
    const parsed = ExecuteCommandArgsSchema.safeParse(args);
    if (!parsed.success) {
      throw new Error(`Invalid arguments for execute_command: ${parsed.error}`);
    }
  
    if (!commandManager.validateCommand(parsed.data.command)) {
      throw new Error(`Command not allowed: ${parsed.data.command}`);
    }
  
    const result = await terminalManager.executeCommand(
      parsed.data.command,
      parsed.data.timeout_ms
    );
  
    return {
      content: [{
        type: "text",
        text: `Command started with PID ${result.pid}\nInitial output:\n${result.output}${
          result.isBlocked ? '\nCommand is still running. Use read_output to get more output.' : ''
        }`
      }],
    };
  }

• src/tools/schemas.ts:4-7 (schema)

  Zod schema for execute_command input: 'command' (string, required), 'timeout_ms' (number, optional).

  export const ExecuteCommandArgsSchema = z.object({
    command: z.string(),
    timeout_ms: z.number().optional(),
  });

• src/server.ts:60-65 (registration)

  Tool registration in the MCP server's listTools response, defining name, description, and input schema.

  {
    name: "execute_command",
    description:
      "Execute a terminal command with timeout. Command will continue running in background if it doesn't complete within timeout.",
    inputSchema: zodToJsonSchema(ExecuteCommandArgsSchema),
  },

• src/server.ts:216-219 (registration)

  Dispatch logic in callTool handler that routes 'execute_command' calls to the handler function after parsing arguments.

  case "execute_command": {
    const parsed = ExecuteCommandArgsSchema.parse(args);
    return executeCommand(parsed);
  }