ai-agentic-hub

AI Agentic Hub

A web GUI to visually build AI agents and workflows using local or cloud LLMs and MCP tool servers. No code required — create agents, connect tools, and build complex agentic workflows from a visual editor.

Features

• Multi-Provider LLM Support — connect to Ollama (local), OpenAI, or Anthropic from the same GUI
• MCP Tool Server Management — add MCP servers, auto-discover tools via MCP SDK
• Agent Builder — create/edit/delete agents with custom system prompts, select any LLM and MCP tools
• Agent Chat — chat with agents, see tool calls and results in real time
• Visual Workflow Editor — drag-and-drop DAG editor (Drawflow) to build agentic workflows
• Orchestrator Loop Pattern — an orchestrator agent dynamically routes tasks to specialist agents and loops until done
• Conditional Edges — route workflow paths based on state (e.g. needs_math == true)
• Shared Workflow State — typed state (LangGraph StateGraph) passes between nodes
• ReAct Agents — agents use LangChain/LangGraph ReAct pattern (reason, act, observe, repeat)
• Mix Providers — use different LLMs for different agents in the same workflow (e.g. orchestrator on Claude, workers on local Ollama)

Prerequisites

• Python 3.12+
• uv (recommended) or pip
• Ollama installed and running with a model (e.g. ollama pull qwen3.5:9b)
• (Optional) OpenAI or Anthropic API key for cloud LLMs
• (Optional) An MCP tool server to connect

Quick Start

1. Clone and install

git clone https://github.com/hasanjawad001/ai-agentic-hub.git
cd ai-agentic-hub
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e .

2. Start the example MCP tool server (optional)

In a separate terminal:

source .venv/bin/activate
start-mcp

This starts 8 example tools on port 3000: add, subtract, multiply, divide, reverse_string, uppercase, lowercase, int_to_string.

3. Start the hub

source .venv/bin/activate
start-hub

Open http://localhost:8000

4. Set up in the GUI

1. LLM Servers — Add your LLM server:
   • Ollama: provider=ollama, url=http://localhost:11434, model=qwen3.5:9b
   • OpenAI: provider=openai, model=gpt-4o, api_key=sk-...
   • Anthropic: provider=anthropic, model=claude-sonnet-4-20250514, api_key=sk-ant-...
2. MCP Servers — Add your MCP server (url: http://localhost:3000) and click Discover Tools
3. Agents — Create agents with system prompts and selected tools
4. Workflows — Create a workflow, open the editor, drag nodes, connect them, and run

Example: Orchestrator Workflow

Build a workflow where an orchestrator agent dynamically routes tasks to specialists:

Start -> Orchestrator <-> Math Agent (loop back)
                      <-> Text Agent (loop back)
         Orchestrator -> End (when done)

Input: compute ((5+5)/(4-2))*3, convert to string, uppercase it, then reverse it

Result: The orchestrator loops 3 times — sends math to the math agent (add, subtract, divide, multiply = 15), then sends text processing to the text agent (uppercase, reverse = "0.51"), then signals done.

All from the visual editor. No code written.

Project Structure

ai-agentic-hub/
├── backend/
│   ├── main.py                 # FastAPI app + page routes
│   ├── database.py             # SQLite setup
│   ├── models.py               # LLMServer, MCPServer, Agent, Workflow
│   ├── api/
│   │   ├── llm_routes.py       # LLM server CRUD + health check
│   │   ├── mcp_routes.py       # MCP server CRUD + tool discovery
│   │   ├── agent_routes.py     # Agent CRUD + chat
│   │   └── workflow_routes.py  # Workflow CRUD + run
│   └── services/
│       ├── llm_service.py      # LangChain LLM client (Ollama/OpenAI/Anthropic)
│       ├── mcp_service.py      # MCP SDK client + LangChain tool conversion
│       ├── agent_service.py    # LangGraph ReAct agent
│       └── workflow_service.py # LangGraph StateGraph workflow engine
├── frontend/templates/         # Jinja2 HTML templates (dark theme)
├── examples/
│   └── test_mcp_server.py      # Example MCP server with 8 tools
├── pyproject.toml
└── LICENSE

Tech Stack

• Backend: Python, FastAPI, SQLModel, SQLite
• Frontend: Jinja2 templates, vanilla JavaScript
• Agent Engine: LangChain + LangGraph (ReAct agents, StateGraph workflows)
• LLM Providers: langchain-ollama, langchain-openai, langchain-anthropic
• Workflow Editor: Drawflow
• MCP Client: mcp Python SDK
• MCP Server (example): FastMCP

License

MIT