Foundry VTT MCP Bridge

import { z } from 'zod'; import dotenv from 'dotenv'; import { getFoundryDataDir, getDefaultComfyUIDir } from './utils/platform.js'; dotenv.config(); const ConfigSchema = z.object({ logLevel: z.enum(['error', 'warn', 'info', 'debug']).default('info'), logFormat: z.enum(['json', 'simple']).default('simple'), enableFileLogging: z.boolean().default(false), logFilePath: z.string().optional(), foundry: z.object({ host: z.string().default('localhost'), port: z.number().min(1024).max(65535).default(31415), namespace: z.string().default('/foundry-mcp'), reconnectAttempts: z.number().min(1).max(20).default(5), reconnectDelay: z.number().min(100).max(30000).default(1000), connectionTimeout: z.number().min(1000).max(60000).default(10000), connectionType: z.enum(['websocket', 'webrtc', 'auto']).default('auto'), protocol: z.enum(['ws', 'wss']).default('ws'), // Legacy, used only for WebSocket mode remoteMode: z.boolean().default(false), dataPath: z.string().optional(), // Custom path for generated maps (remote mode) rejectUnauthorized: z.boolean().default(true), // TLS certificate validation // WebRTC configuration webrtc: z.object({ stunServers: z.array(z.string()).default([ 'stun:stun.l.google.com:19302', 'stun:stun1.l.google.com:19302' ]), // Future: TURN servers support // turnServers: z.array(z.object({ // urls: z.string(), // username: z.string().optional(), // credential: z.string().optional() // })).optional() }).default({ stunServers: ['stun:stun.l.google.com:19302', 'stun:stun1.l.google.com:19302'] }) }), comfyui: z.object({ // ComfyUI always runs locally on the same machine as the MCP server port: z.number().min(1024).max(65535).default(31411), installPath: z.string(), // No default here - set in rawConfig host: z.string().default('127.0.0.1'), pythonCommand: z.string().default('python/python.exe'), // Will be platform-specific }), server: z.object({ name: z.string().default('foundry-mcp-server'), version: z.string().default('0.4.17'), }), }); export type Config = z.infer<typeof ConfigSchema>; const rawConfig = { logLevel: process.env.LOG_LEVEL || 'warn', logFormat: process.env.LOG_FORMAT || 'simple', enableFileLogging: process.env.ENABLE_FILE_LOGGING === 'true', logFilePath: process.env.LOG_FILE_PATH, foundry: { host: process.env.FOUNDRY_HOST || 'localhost', port: parseInt(process.env.FOUNDRY_PORT || '31415', 10), namespace: process.env.FOUNDRY_NAMESPACE || '/foundry-mcp', reconnectAttempts: parseInt(process.env.FOUNDRY_RECONNECT_ATTEMPTS || '5', 10), reconnectDelay: parseInt(process.env.FOUNDRY_RECONNECT_DELAY || '1000', 10), connectionTimeout: parseInt(process.env.FOUNDRY_CONNECTION_TIMEOUT || '10000', 10), connectionType: (process.env.FOUNDRY_CONNECTION_TYPE || 'auto') as 'websocket' | 'webrtc' | 'auto', protocol: (process.env.FOUNDRY_PROTOCOL || 'ws') as 'ws' | 'wss', remoteMode: process.env.FOUNDRY_REMOTE_MODE === 'true', dataPath: process.env.FOUNDRY_DATA_PATH, rejectUnauthorized: process.env.FOUNDRY_REJECT_UNAUTHORIZED !== 'false', webrtc: { stunServers: process.env.FOUNDRY_STUN_SERVERS ? process.env.FOUNDRY_STUN_SERVERS.split(',') : ['stun:stun.l.google.com:19302', 'stun:stun1.l.google.com:19302'] } }, comfyui: { // ComfyUI always runs locally on the same machine as the MCP server (localhost:31411) port: parseInt(process.env.COMFYUI_PORT || '31411', 10), installPath: process.env.COMFYUI_INSTALL_PATH || getDefaultComfyUIDir(), host: process.env.COMFYUI_HOST || '127.0.0.1', pythonCommand: process.env.COMFYUI_PYTHON_COMMAND || 'python/python.exe' }, server: { name: process.env.SERVER_NAME || 'foundry-mcp-server', version: process.env.SERVER_VERSION || '1.0.0', }, }; export const config = ConfigSchema.parse(rawConfig); /** * WebRTC Protocol Constants * * SCTP (Stream Control Transmission Protocol) limits for WebRTC data channels */ export const WEBRTC_CONSTANTS = { /** * SCTP maxMessageSize limit - hard limit imposed by WebRTC specification * Messages exceeding this size will fail with "OperationError: Failure to send data" */ MAX_MESSAGE_SIZE: 65536, // 64KB in bytes /** * Safe chunk size threshold for splitting large messages * Set below MAX_MESSAGE_SIZE to account for: * - JSON stringification overhead (~1-2KB for chunk metadata) * - String escaping (quotes, backslashes, unicode) can increase size 5-20% * - Safety buffer to prevent edge cases * * Testing showed 50KB provides reliable chunking with ~14KB headroom */ CHUNK_SIZE: 50 * 1024, // 50KB in bytes /** * Timeout for incomplete chunked messages (milliseconds) * After this time, pending chunks are cleaned up to prevent memory leaks * * Network issues or client disconnects can leave incomplete messages * Set to 30 seconds - longer than typical query timeout (10s) but prevents indefinite storage */ CHUNK_TIMEOUT_MS: 30000, // 30 seconds /** * Maximum chunks allowed per message * Security limit to prevent "chunk bomb" attacks where malicious clients * send huge totalChunks values to trigger memory allocation attacks * * 1000 chunks * 50KB = 50MB maximum message size * This is far larger than any legitimate Foundry VTT data */ MAX_CHUNKS_PER_MESSAGE: 1000, /** * Interval for cleanup of timed-out chunks (milliseconds) * Background task runs periodically to remove incomplete messages */ CHUNK_CLEANUP_INTERVAL_MS: 10000, // 10 seconds } as const;