AgentFlow 🚀

AgentFlow is a next-generation, premium agentic workflow system built on the Model Context Protocol (MCP). It transforms the way AI agents handle complex development tasks by bridging the gap between raw LLM reasoning and structured execution.

✨ The Vision: "Vibe Coding" Refined

AgentFlow is designed for the modern "Vibe Coder"—developers who prioritize high-level intent, rapid iteration, and stunning visual feedback.

• 🧠 Integrated Thought Chain: Native Support for complex multi-step reasoning (plan → analyze → reflect).
• 🎨 State-of-the-Art Dashboard: A premium Glassmorphism interface with Dynamic Background Gradients, fluid animations (15s ease infinite shift), and meticulous Light/Dark themes.
• 🎮 Real-Time Pulse: Live updates via SSE (Server-Sent Events). Watch your agent create, update, and reorder tasks in real-time.
• ⚛️ Dual Persistence: Robust storage support via SQLite for local portability or Supabase for enterprise-scale persistence.
• 📁 Intelligent Project Context: Auto-detects projects based on workspace paths or Git remote URLs, mapping local work to stable project identities.

🧭 The Core Workflow

AgentFlow promotes a high-integrity "Think-Then-Do" pipeline:

1. 💡 Plan Idea: Draft initial concepts and design tokens.
2. 🔍 Analyze: Deep dive into technical requirements and blockers.
3. 🤔 Reflect: Self-critique the plan to catch edge cases before a single line of code is written.
4. 🧩 Split: Automatically decompose the refined plan into a Topological DAG of tasks.
5. ▶️ Execute: Move through tasks with built-in Verification Loops ensuring criteria are met.

⚙️ Technical Stack

• Backend: Node.js + TypeScript (High-performance ESM).
• MCP Implementation: @modelcontextprotocol/sdk for seamless tool-calling integration.
• Persistence: SQLite (Local) / Supabase (Remote) via shared DatabaseAdapter.
• Dashboard: Vanilla JS + CSS (Custom Tokens) + D3.js for real-time dependency graph visualization.
• Communication: HTTP REST API + Server-Sent Events (SSE) for zero-latency UI updates.

🛠️ Advanced Toolset

🏗️ Reasoning Engine

📋 Task Orchestration

� Quick Start

🔽 Via Smithery (Recommended)

npx -y @smithery/cli install agent-flow --client claude

🔽 Manual Setup

git clone https://github.com/farazpawle/agent-flow.git
cd agent-flow
npm install
npm run build

🔌 Configuration

AgentFlow is optimized for Zero-Config starts, but uses a .env file for advanced customization.

1. Setup Environment

In your project root (where you run the agent), create a .env file:

# Example .env configuration
DATA_DIR="C:/MyProject/agent-flow-data" # REQUIRED: Absolute path
DB_TYPE="sqlite"                        # Options: sqlite, supabase
ENABLE_GUI="true"                      # Enable the visual dashboard
ENABLE_DETAILED_MODE="true"             # Record conversation history

2. MCP Client Registration

Register the server in your MCP client (Cursor, Claude Desktop, etc.). Since AgentFlow auto-loads your .env, the configuration is minimal:

Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "agent-flow": {
      "command": "npx",
      "args": ["-y", "agent-flow"]
    }
  }
}

💡 Pro Tip: You can still override any .env setting by passing it directly in the env block of your client JSON if needed.

☁️ Supabase Cloud Setup

If you want to use Supabase instead of SQLite:

1. Initialize Tables: Copy the content of scripts/supabase-schema.sql and run it in your Supabase SQL Editor.
2. Verify Setup: Run npm run supabase:check to ensure all tables are correctly created and accessible.
3. Migrate local data: If you have local SQLite data, run npm run supabase:migrate to push it to the cloud.
4. Update .env:

   DB_TYPE="supabase"
   SUPABASE_URL="https://your-project.supabase.co"
   SUPABASE_SERVICE_KEY="your-service-role-key"
   

📄 License

MIT License. Created with ❤️ for the Vibe Coding community.