Ai Architect

AI Architect

Natural language → valid ComfyUI workflow JSON

---

Screenshots

---

What it does

Describe an image generation pipeline in plain English; AI Architect produces a valid ComfyUI workflow JSON you can load or execute immediately. It supports multiple AI providers (Claude, GPT-4o, Gemini, OpenRouter) and builds context-aware prompts from the live node registry of your running ComfyUI instance. Generated workflows go through multi-stage validation and are automatically corrected when the model produces invalid node connections or missing inputs.

---

Features

• Multi-provider AI — 12+ models across Anthropic, OpenAI, Google, and OpenRouter; per-model token budgeting
• Visual node graph — ReactFlow canvas with auto-layout, type-checked edge connections, and inline editing
• Live ComfyUI integration — /object_info sync for node discovery, WebSocket execution pipeline, error auto-correction loop
• Context-aware prompting — system prompt assembled from live node registry + installed checkpoints/LoRAs at generation time
• MCP server — exposes ComfyUI operations as tools consumable by any MCP-compatible AI client
• Zero-backend architecture — fully client-side; AI calls go browser-direct, state persisted in localStorage

---

Stack

---

Status

| Component | Status | |---|---| | Frontend (browser) | Complete — functional in browser dev mode | | Electron desktop shell | Complete — window opens with full API access | | Test suite | Not present |

---

Setup

Prerequisites: Node 20+, pnpm, a running ComfyUI instance

cd webapp
pnpm install
cp .env.example .env          # fill in your AI provider keys
pnpm dev                      # browser — http://localhost:5173
pnpm electron:dev             # desktop (Electron)

Set VITE_COMFYUI_URL in .env to point at your ComfyUI instance (default http://127.0.0.1:8188).

For local path overrides (ComfyUI root, Python exe, etc.):

cp .env.secret.example .env.secret   # fill in absolute paths, never committed

---

MCP Server

cd mcp-server
pnpm install
pnpm start

Add the server to your MCP client config (e.g. .mcp.json) to expose ComfyUI tools (queue_prompt, get_history, get_models, etc.) to your AI client.

---

Project Layout

webapp/          React + Vite frontend + Electron shell
mcp-server/      TypeScript MCP server
scripts/         Bootstrap and check scripts (PowerShell)
docs/            Screenshots and internal docs

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fullydigital-design
• Source: fullydigital-design/ai-architect
• License: MIT

Install and usage instructions live in the source repository linked above.