🚀 AgentFoundry

Forge, orchestrate, and extend intelligent multi-agent systems with RAG, MCP, and A2A communication.

---

🧠 What is AgentFoundry?

AgentFoundry is a developer-first framework for building and experimenting with collaborative multi-agent systems. It provides a sandbox and a framework, giving developers both flexibility and power to:

• Build domain-specific intelligent agents,
• Connect them through shared memory and A2A protocols,
• Integrate external capabilities using MCP servers and
• Extend the system without rewriting the core.

AgentFoundry is designed for researchers, ML engineers, and AI developers who want to move beyond single-agent pipelines and explore true agentic cooperation.

---

🚀 Why AgentFoundry?

AgentFoundry is designed for developers who want full control over their agentic systems. It provides a modular, transparent, and hackable framework that allows you to:

• Prototype your own AI orchestration logic
• Plug in custom tools or APIs dynamically
• Build domain-specific AI agents (e.g., medical assistants, research copilots, or autonomous workflow bots)
• Experiment with A2A (Agent-to-Agent) collaboration and distributed reasoning

Each component in AgentFoundry is replaceable, extensible, and traceable. You have full control over the planning logic, message routing, RAG retrieval, and MCP-based tool integration. This means you can easily customize and experiment with different architectures and use cases.

🎯 Key Features

• Planner Agent: Breaks down complex tasks into subtasks.
• Worker Agents: Execute the subtasks.
• Verifier Agent: Validates outputs for correctness.
• RAG Integration: Dynamic retrieval via agents.
• MCP Server Integration: Enables external tools.
• Memory Layer: Central memory module managing shared state and data access across agents.
• Extensibility: Easily add more tools (APIs, DBs, system commands) via MCP.
• Centralized or Decentralized control: Choose between a central orchestrator or decentralized A2A communication between agents.

---

Agent Flow (general):

1. User Input: Direct command via interactive CLI
2. Planner: Analyzes request and creates task plan
3. Worker: Executes tasks using available tools:
   • MCP Tools: file_search, read_file, save_file(mock tools present in this repo)
   • RAG Tool: Knowledge base retrieval when needed
4. Verifier: Validates and refines outputs
5. Real-time Response: Live feedback and results to user

---

🛠️ Tech Stack

• Python
• LangChain – agent orchestration
• Google Gemini LLM – reasoning & generation
• Chroma / FAISS – vector store for RAG
• FastMCP – lightweight MCP server
• MCP Toolkit for LangChain – dynamic tool integration
• Mem0 - memory layer for agents

---

📂 Project Structure

agentfoundry/
├── scripts/
│   ├── a2a_network.py
│   └── orchestrator.py
├── src/
│   ├── agents/
│   │   ├── planner.py
│   │   ├── worker.py
│   │   └── verifier.py
│   ├── core/
│   │   ├── central.py
│   │   └── messaging.py
│   ├── RAG/
│   │   ├── load_docs.py
│   │   ├── retriever.py
│   │   ├── rag_tool.py
│   │   └── vector_store.py
│   ├── MCP/
│   │   ├── MCP_servers.py
│   │   └── mcp_tools_adapter.py
│   ├── memory/
│   │   └── memory_manager.py
│   └── config/
│       ├── __init__.py
│       └── settings.py
├── rag_db/
├── pyproject.toml
├── README.md
├── .env
├── .gitignore
└── requirements.txt

🚀 Getting Started

1. Clone the Repository

git clone https://github.com/your-username/AgentFoundry.git
cd AgentFoundry

2. Install (editable)

pip install -e .

3. Run MCP Server (only when running over a different port)

python -m MCP.MCP_servers

4. Run the Orchestrator

python scripts/orchestrator.py

OR

5. Run the A2A Network

python scripts/a2a_network.py

Environment Setup

• .env: define GOOGLE_API_KEY or GEMINI_API_KEY, plus any other provider keys you enable.
• Paths: override RAG_DB_PATH (defaults to rag_db/) and MEM0_CHROMA_PATH (defaults to .mem0_chroma/) if you want custom storage locations.
• MCP transport: switch from the default stdio by setting MCP_TRANSPORT, MCP_SERVER_HOST, and MCP_SERVER_PORT.

📘 Usage Examples

See USAGE.md for concise terminal transcripts (orchestrator, A2A) and quick tuning tips. Also read the documentation in scripts directory to know more about agent coummunication.

🤝 Contribution

AgentFoundry is developer-facing — contributions are welcome

Feel free to open issues or submit pull requests! its a test