architect

Architect MCP is an AI agent workshop that lets your agents dynamically create and deploy their own tools within a secure sandbox. The server provides a built-in workflow where agents can write JavaScript tools, request user-approved permissions, and execute those tools in isolation. Key features include a real-time web dashboard (accessible on port 3001 by default) for managing active tools, execution logs, failures, and secrets, as well as a persistent SQLite storage layer for tools, run logs, and execution state. Agents can also schedule tools to run on cron-like intervals, and browse or publish tools to a global marketplace using a GitHub token. This makes Architect a flexible, extendable platform where your agents can go from problem-solving to tool-building in a secure, auditable environment. To get started, run the server locally or via Docker, then open http://localhost:3001 to monitor and manage tools as they are created and executed by your agents.

Prerequisites:

• Node.js (recommended LTS) and npm installed on your system
• Docker and Docker Compose installed if you prefer the Docker setup

Install and run locally:

1. Clone the repository

2. Install dependencies npm install

3. Build (if applicable) npm run build

4. Start the server npm start

5. Access the dashboard at http://localhost:3001

Docker setup (recommended for isolation):

1. Ensure Docker and Docker Compose are installed
2. From the project directory, start with Docker Compose docker compose up -d
3. The dashboard will be available at http://localhost:3001 and data/tools will persist in Docker volumes

Prereq notes:

• The server expects a Node.js environment and will run tools inside a sandboxed context.
• If you modify environment-sensitive settings (ports, database paths, or secrets), update your environment variables accordingly.

Tips and common considerations:

• The built-in sandbox ensures tools run with strict permission checks. You must explicitly approve network or filesystem access for each tool before execution.
• The dashboard on port 3001 lets you review tool requests, grant granular permissions, and monitor real-time logs.
• Tools and runs are stored in a SQLite-backed storage layer; for production, consider backing store options or persistent Docker volumes.
• You can leverage the Global Marketplace to browse, install, or publish tools using a GitHub token.
• Cron scheduling allows agents to run tools on schedules or as part of continuous pipelines.
• If you encounter port conflicts, adjust the PORT environment variable or Docker Compose configuration.
• Environment variables you might configure include PORT, DB_PATH, and LOG_LEVEL for debugging.